View HSP50216_3208350.PDF datasheet online --- IC-ON-LINE

Datasheet File OCR Text:

1 tm file number 4557.3 caution: these devices are sensitive to electrosta tic discharge; follow proper ic handling procedures. 1-888-intersil or 321-724-7143 for more information contact: juan ga rcia - 321-729-5883 | copyright ? intersil americas inc. 2001, all rights reserved HSP50216 four-channel programmable digital downconverter the HSP50216 quad programmable digital downconverter (qpdc) is designed for high dynamic range applications such as cellular basestations where multiple channel processing is required in a small physical space. the qpdc combines into a single package, a set of four channels which include: digital mixers, a quadrature carrier nco, digital filters, a resampling filter, a cartesian-to-polar coordinate converter and an agc loop. the HSP50216 accepts four channel s of 16-bit real digitized if samples which are mixed wit h local quadrature sinusoids. each channel carrier nco frequency is set independently by the microprocessor. the output of the mixers are filtered with a cic and fir filters, with a variety of decimation options. gain adjustment is provided on the filtered signal. the digital agc provides a gain adjust range of up to 96db with programmable thresholds and slew rates. a cartesian to polar coordinate converter provides magnitude and phase outputs. a frequency discriminator provides a frequency output via the fir filter. se lectable outputs include i samples, q samples, magnitude, phase, frequency and agc gain. the output resolution is selectable from 4-bit fixed point to 32-bit floating point. the maximum output bandwidth achievable using a single channel is at least 1mhz. features ? up to 70msps input ? four independently programmable downconverter channels in a single package ? four parallel 16-bit inputs - fixed or floating point format ? 32-bit programmable carrier nco with > 115db sfdr ? 110db fir out of band attenuation ? decimation from 8 to >65536 ? 24-bit intern al data path ? digital agc with up to 96db of gain range ? filter functions - 1 to 5 stage cic filter - halfband decimation and interpolation fir filter - programmable fir filter - resampling fir filter ? cascadable filtering for additional bandwidth ? four independent serial outputs ? 3.3v operation applications ? narrow-band tdma through is-95 cdma digital software radio and basestation receivers ? wide-band applications: w-cdma and umts digital software radio and basestation receivers ordering information part number temp range ( o c) package pkg. no HSP50216ki -40 to 85 196 ld bga v196.12x12 data sheet april 2001
2 block diagram p interface a(15:0) clk level nco / mixer / cic input select, format, demux detector p test register p mode b(15:0) c(15:0) d(15:0) reset synci synco channel 0 p(15:0) add(2:0) wr ce sclk rd or dstrb or rd / wr intrpt synca sdia sd2a output select, format, serialize input select, format, demux bus i q nco / mixer / cic channel 1 input select, format, demux i q nco / mixer / cic channel 2 input select, format, demux i q nco / mixer / cic channel 3 input select, format, demux i q routing syncb sdib sd2b syncc sdic sd2c syncd sdid sd2d enia enib enic enid fir filters, agc, cartesian-to-polar coordinate converter fir filters, agc, cartesian-to-polar coordinate converter fir filters, agc, cartesian-to-polar coordinate converter fir filters, agc, cartesian-to-polar coordinate converter HSP50216
3 pinout 196 lead bga top view k j h g f e d c b a 123456789 11 10 l m n vcc d9 gnd vcc 12 d7 13 14 c14 c10 c8 gnd vcc gnd d11 enid d13 b1 d15 d3d1d0 b0 c12c6c4c2c0 b3 wr b2 b5 gnd p0 vcc b7 p2 gnd clk gnd p4 vcc b9 vcc p6 gnd b11 gnd p8 vcc b13 p10 gnd b15 p12 sd2c sd2d sd2b sd1b intrpt p15 enia a12 a14 sd2a sd1c gnd vcc gnd add0 a8 a10 gnd vcc sclk syncc syncb synca syncd synci synco a7 a9 a11 a13 a15 sd1a p d12 d10 d14 c13 enic d8 d6 d4 c11c9c7c5c3c1 power pin ground pin signal pin thermal ball nc (no connection) b12 b14 b10 gnd b8 b6 vcc p3 p5 p7 a5 a6 a3 vcc sd1d add1 a4 a2 a1 p14 a0 p13 reset vcc p11 p9 b4 p1 enib rd ce c15 d5 d2 add2 p mode HSP50216
4 pin descriptions name type description power supply vcc - positive power supply voltage, 3.3v 0.15 gnd - ground, 0v. inputs a(15:0) i parallel data input bus a. samp led on the rising edge of clock when enia is active (low). b(15:0) i parallel data input bus b. samp led on the rising edge of clock when enib is active (low). c(15:0) i parallel data input bus c. samp led on the rising edge of clock when enic is active (low). d15 i parallel data input d15 or tuner channel a cof. d14 i parallel data input d14 or tuner channel a cofsync. d13 i parallel data input d13 or tuner channel a sof. d12 i parallel data input d12 or tuner channel a sofsync. d11 i parallel data input d11 or tuner channel b cof. d10 i parallel data input d10 or tuner channel b cofsync. d9 i parallel data input d9 or tuner channel b sof. d8 i parallel data input d8 or tuner channel b sofsync. d7 i parallel data input d7 or tuner channel c cof. d6 i parallel data input d6 or tuner channel c cofsync. d5 i parallel data input d5 or tuner channel c sof. d4 i parallel data input d4 or tuner channel c sofsync. d3 i parallel data input d3 or tuner channel d cof. d2 i parallel data input d2 or tuner channel d cofsync. d1 i parallel data input d1 or tuner channel d sof. d0 i parallel data input d0 or tuner channel d sofsync. enia i input enable for parallel data input bus a. active low. this pin enables the input to the part in one of two modes, gated or interpolated. in gated mode, one sample is taken per clk when eni is asserted. enib i input enable for parallel data input bus b. active low. this pin enables the input to the part in one of two modes, gated or interpolated. in gated mode, one sample is taken per clk when eni is asserted. enic i input enable for parallel data input bus c. active low. this pin enables the input to the part in one of two modes, gated or interpolated. in gated mode, one sample is taken per clk when eni is asserted. enid i input enable for parallel data input bus d. active low. this pin enables the input to the part in one of two modes, gated or interpolated. in gated mode, one sample is taken per clk when eni is asserted. control clk i input clock. all processing in t he HSP50216 occurs on the rising edge of clk. synci i synchronization input signal. used to align the proc essing with an external event or with other HSP50216 devices. synci can update the carrier nco, reset dec imation counters, restart the filter compute engine, and restart the output section among other functions. for most of the functional blocks, the response to synci is programmable and can be enabled or disabled. synco o synchronization output signal . the processing of multiple h sp50216 devices can be synchronized by tying the synco from one HSP50216 device (the master ) to the synci of all the HSP50216 devices (the master and slaves). reset i reset signal. active low. asserting reset will halt all processing and set certain registers to default values. HSP50216
5 outputs sd1a o serial data output 1a. a serial data stream output which can be programmed to consist of i1, q1, i2, q2, magnitude, phase, frequency (d /dt), agc gain, and/or zeros. in addition, data outputs from channels 0, 1, 2 and 3 can be multiplexed into a common serial output data stream. information can be sequenced in a programmable order. see serial data output formatter section and microprocessor interface section. sd2a o serial data output 2a. this output is provided as an auxiliary output for serial data output 1a to route data to a second destination or to output two words at a time for higher sample rates. sd2a has the same programmability as sd1a except that fl oating point format is not available. see serial data output formatter section and microprocessor interface section. sd1b o serial data output 1b. see description for sd1a. sd2b o serial data output 2b. see description for sd2a. sd1c o serial data output 1c. see description for sd1a. sd2c o serial data output 2c. see description for sd2a. sd1d o serial data output 1d. see description for sd1a. sd2d o serial data output 2d. see description for sd2a. sclk o serial output clock. can be programmed to be at 1, 1/2, 1/4, 1/8, or 1/16 times the clock frequency. the polarity of sclk is programmable. synca o serial data output 1a sync signal. this signal is us ed to indicate the start of a data word and/or frame of data. the polarity and position of synca is programmable. syncb o serial data output 1b sync signal. this signal is us ed to indicate the start of a data word and/or frame of data. the polarity and position of syncb is programmable. syncc o serial data output 1c sync signal. this signal is used to indicate the start of a data word and/or frame of data. the polarity and position of syncc is programmable. syncd o serial data output 1d sync signal. this signal is used to indicate the start of a data word and/or frame of data. the polarity and position of syncd is programmable. microprocessor interface p(15:0) i/o microprocessor interface data bus. see microprocessor interface section . p15 is the msb. add(2:0) i microprocessor interface address bus. add2 is the msb. see microprocessor interface section . note: add2 is not used but designated for future expansion. wr or dstrb i microprocessor interface write or data strobe signal. when the microprocessor interface mode control, p mode, is a low data transfers (from either p(15:0) to the internal write holding register or from the internal write holding register to the target register specified) occur on the low to high transition of wr when ce is asserted (low). when the p mode control is high this input fu nctions as a data read/write strobe. in this mode with rd/wr low data transfers (from either p(15:0) to the internal write holding register or from the internal write holding register to the target register specified) occur on the low to high transition of data strobe. with rd/wr high the data from the address specified is placed on p(15:0) when data strobe is low. see microprocessor interface section . rd or rd/wr i microprocessor interface read or read/write signal. when the microprocessor interface mode control, p mode, is a low the data from the address specified is placed on p(15:0) when rd is asserted (low) and ce is asserted (low). when the p mode control is high this i nput functions as a read/write control input. data is read from p(15:0) when high or written to the appropriate register when low. see microprocessor interface section . p mode i microprocessor interface mode control. this pin is used to select the read/write mode for the microprocessor interface. internally pulled down. see microprocessor interface section . ce i microprocessor interface chip select. active low. this pin has the same timing as the address pins. intrpt o microprocessor interrupt signal. asserted for a prog rammable number of clock cycles when new data is available on the selected channel. pin descriptions (continued) name type description HSP50216
6 functional description the HSP50216 is a four channel digital receiver integrated circuit offering exceptional dynamic range and flexibility. each of the four channels co nsists of a front-end nco, digital mixer, and cic-filter block and a back-end fir, agc and cartesian to polar coordinate-conversion block. the parameters for the four channels are independently programmable. four parallel data input busses (a(15:0), b(15:0), c(15:0) and d(15:0)) and four pairs of serial data outputs (sdxa, sdxb, sdxc, and sdxd; x = 1 or 2) are provided. each input can be co nnected to any or all of the internal signal processing channels, channels 0, 1, 2 and 3. the output of each channel can be routed to any of the serial outputs. outputs from more than one channel can be multiplexed through a common output if the channels are synchronized. the four channels share a common input clock and a common serial out put clock, but the output sample rates can be synchronous or asynchronous. bus multiplexers between the front end and back end sections provide flexible routing between channels for cascading back-end filters or for routing one front end to multiple back ends for polyphase filtering or systolic arrays (to provide wider bandwidth filtering). a level detector is provided to monitor the signal level on any of the parallel data input busses, facilitating microproce ssor control of gain blocks prior to an a/d converter. each front end nco/digital mixer/cic filter section includes a quadrature numerically controll ed oscillator (nco), digital mixer, barrel shifter and a ca scaded-integrato r-comb filter (cic). the nco has a 32-bit frequency control word for 16.3mhz tuning resolution at an input sample rate of 70msps. the sfdr of the nco is >115db. the barrel shifter provides a gain of between 2 -45 and 2 -14 to prevent overflow in the cic. the cic filter order is programmable between 1 and 5 and the cic decimation factor can be programmed from 4 to 512 for 5 th order, 2048 for 4 th order, 32768 for 3 rd order, or 65536 for 1 st or 2 nd order filters. each channel back end section includes an fir processing block, an agc and a cartesian-to-polar coordinate converter. the fir processing block is a flexible filter compute engine that can compute a single fir or a set of cascaded decimating filters. a single filter in a chain can have up to 256 taps and the total number of taps in a set of filters can be up to 384 provi ded that the decimation is sufficient. the HSP50216 calculates 2 taps per clock (on each channel) for symmetric filters, generally making decimation the limiting factor for the number of taps available. the filter compute engine supports a variety of filter types including deci mation, interpolation and resampling filters. the coefficients for the programmable digital filters are 22 bits wide. coefficients are provided in rom for several halfband filter responses and for a resampler. the agc section can provide up to 96db of either fixed or automatic gain control. for automatic gain control, two settling modes and two sets of loop gains are provided. separate attack and decay slew rates are provided for each loop gain. programmable limits allow the user to select a gain range less than 96db. the outputs of the cartesian-to-polar coordinate conversion block, used by the agc loop, are also provided as outputs to the user for am and fm demodulation. the HSP50216 supports both fixed and floating point parallel data input modes. the floating point modes support gain ranging a/d converters. gated, interpolated and multiplexed data input modes are supported. the serial data output word width for each data type can be programmed to one of ten output bit widths from 4-bit fixed point through 32- bit ieee 754 fl oating point. the HSP50216 is programmed through a 16-bit microprocessor interface. the output data can also be read via the microprocessor interface for all channels that are synchronized. the HSP50216 is specified to operate to a maximum clock rate of 70m sps over the industrial temperature range (-40 o c to 85 o c). the power supply voltage range is 3.3v 0.15v. the i/os are not 5v tolerant. HSP50216
7 input select/format block each front end block and the level detector block contains an input select/format block. a functional block diagram is provided in the above figure. t he input source can be any of the four parallel input busses (see microprocessor interface section table 3, iwa *000h) or a test register loaded via the processor bus (see microprocessor interface section table 42, gwa f807h). the input to the part can operate in a gated or interpolated mode. each input channel has an input enable (enix ,x=a, b, c or d). in the gated mode, one input sample is processed per clock that the enix signal is asserted (low). processing is disabled when enix is high. the enix signal is pipelined through the part to minimize delay (latency). in the interpolated mode, the input is zeroed when the enix signal is high, but processing inside the part continues. this mode inserts zeros between the data samples, interpolating the input data stream up to the clo ck rate. on reset, the part is set to gated mode and the input enables are disabled. the inputs are enabled by the first synci signal. the input section can select one channel from a multiplexed data stream of up to 8 channels. the input enable is delayed by 0 to 7 clock cycles to enab le a selection register. the register following the selection register is enabled by the non-delayed input enable to realign the processing of the channels. the one-clock-wide in put enable must align with the data for the first channel. the desired channel is then selected by programming the delay. a delay of zero selects the first channel, a delay of 1 selects the second, etc. the parallel input busses are 16 bits wide. the input format may be twos complement or offset binary format. a floating point mode is also supported. the floating point modes and the mapping of the parallel 16-bit input format is discussed below. a(15:0) enia b(15:0) enib c(15:0) enic d(15:0) enid note: eni* signals are active high (inverted at the i/o pad) testenbit testenstrb mux mux external data input select (iwa *000 - 14:13 or gwa f804 - 14:13) mux p test register (gwa f807 - 15:0) (iwa *000 - 11 (gwa f808) or gwa f804 - 11) test eni select (iwa *000 - 12 external/test select (iwa *000 - 15 or gwa f804 - 15) or gwa f804 - 12) 15:0 15:0 testen eni 15:0 en offset binary or two?s complement (iwa *000 - 10 or gwa f804 - 10) format floating point to fixed point mux 11/3, 12/3, 13/3, 14/2 (iwa *000 - 8:7 or gwa f804 - 8:7) r e g programmable delay de-multiplex control (0-7) (iwa *000 - 6:4 or gwa f8o4 - 6:4) 15:0 fixed point or floating point (iwa *000 - 9 or gwa f804 - 9) data to nco / mixer interpolated/gated mode (iwa *000 - 3 or gwa f804 - 3) input enable hold off (enabled by synci) (gwa f802 - 30) data sample enable carrier offset frequency (cof) cof sync enable cof resampler offset frequency (sof) sof sync enable sof enable pn (iwa *000 - 0) pn cof to carrier nco/mixer cof sync to carrier nco/mixer sof to resampler nco sof sync to resampler nco pn to carrier nco/mixer (1wa *000 - 2) (iwa *000 - 1) or level detector HSP50216
8 floating point input mode bit mapping the input bit weighting for fixed point inputs on busses a, b, c, and d is: bit 15 (msb): 2 0 , bit 14: 2 -1 , bit 13: 2 -2 , ..., bit 0: 2 -15 . for floating point modes, the leas t significant 2 or 3 bits are used as exponent bits (see floating point input mode bit mapping tables). the difference between the four floating point modes with three exponent bits is where the exponent saturates. floating point input m ode bit mapping tables a(15:0), b(15:0), c(15:0) or d(15:0): 1514 131211109876 5 4 3 2 1 0 2 0 2 -1 2 -2 2 -3 2 -4 2 -5 2 -6 2 -7 2 -8 2 -9 2 -10 2 -11 2 -12 2 -13 /(exp2) (exp1) (exp0) 11-bit mode: 11 to 13-bit mantissa , 3-bit exponent, 30db exponent range exponent gain (db) pin bit weig hting to 16-bit input mapping 000 0 x15 x15 x15 x15 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 001 6 x15 x15 x15 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 010 12 x15x15x15x15x14x13x12x11x10x9x8x7x6x5x4x3 011 18 x15 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 0 100 24 x15x15x14x13x12x11x10x9x8x7x6x5x4x3 0 0 101 (note 1) 30 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 0 0 0 notes: 1. or 110 or 111, the exponent input saturates at 10. 2. ?xnn? = input a, b, c, or d bit nn. 12-bit mode: 12 to 13-bit mantissa , 3-bit exponent, 24db exponent range exponent gain (db) pin bit weighting to 16-bit input mapping 000 0 x15 x15 x15 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 001 6 x15 x15 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 010 12 x15 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 0 011 18 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 0 0 100 (note 3) 24 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 0 0 0 note: 3. or 101, 110, or 111, the exponent input saturates at 100. 13-bit mode: 13-bit mantissa, 3- bit exponent, 18db exponent range exponent gain (db) pin bit weighting to 16-bit input mapping 000 0 x15 x15 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 001 6 x15 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 0 010 12 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 0 0 011 (note 4) 18 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 0 0 0 note: 4. or 100, 101, 110, or 111, the exponent input saturates at 011. 14-bit mode: 14-bit mantissa, 2- bit exponent, 12db exponent range exponent gain (db) pin bit weighting to 16-bit input mapping 00 0 x15 x15 x15 x14 x13 x12 x11 x10 x9x8x7x6x5x4x3x2 01 6 x15 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 x2 0 10 (note 5) 12 x15 x14 x13 x12 x11 x10 x9 x8 x7 x6 x5 x4 x3 x2 0 0 note: 5. or 11, the exponent input saturates at 10. HSP50216
9 level detector an input level detector is provided to monitor the signal level on any of the input busses. which input bus, the input format, and the level detection type are programmble (see microprocessor interface section tables 39, 40 and 41, gwa?s f804h, f805h and f806h). this signal level represents the wideband signal from the a/d and is useful for controlling gain / attenuation blocks ahead of the converter. the supported monitoring modes are: integrated magnitude (like the hsp50214 w/o the threshold), leaky integration (y n =x n xa+y n-1 x (1-a)) where a = 1, 2 -8 , 2 -12 , or 2 -16 (see gwa = f805h), and peak detection. the measurement interval can be programmed from 2 to 65537 samples (or continuous for the leaky integr ator and peak detect cases). the output is 32 bits and is read via the p interface. nco/mixer after the input select/format section, the samples are multiplied by quadrature sine wa ve samples from the carrier nco. the nco has a 32-bit fr equency control, providing sub-hertz resolution at the maximum clock rate. the quadrature sinusoids have except ional purity. the purity of the nco should not be the determining factor for the receiver dynamic range performance. the phase quantization to the sine/cosine generator is 24 bits and the amplitude quantization is 19 bits. the carrier nco center frequency is loaded via the p bus. the center frequency control is double buffered - the input is loaded into a center frequency holding register via the p interface. the data is then trans ferred from the holding register to the active register by a write to a address iwa *006h or by a synci signal, if loading via synci is enabled. to synchronize multiple channels, the carrier nco phase accumulator feedback can be zeroed on loading to restart all of the ncos at the same phase. a serial offset frequency input is also available for each channel through the d(15:0) parallel data input bus (if that bus is not needed for data input). this is legacy support for hsp50210 type tracking signals. see iwa=*000 and *004 for carrier offset frequency parameters. after the mixers, a pn (pseudonoise) signal can be added to the data. this feature is provided for test and to digitally reduce the input sensitivity and adjust the receiver range (sensitivity). the effect is the same as increasing the noise figure of the receiver, reducing its sensitivity and overall dynamic range. for testing, the pn generator provides a wideband signal which may be used to verify the frequency response of a filter. the one bit pn data is scaled by a 16-bit programmable scale factor. the overall range for the pn is 0 to 1/4 full scale (see iwa = *001h). a gain of 0 disables the pn input. the pn value is formed as where s is the pn generator output bit (treated as a sign bit) and the 16 x?s refer to the pn gain register iwa = *001h. the minimum, non-zero, pn value is 2 -18 of full scale (-108dbfs) on each axis (-105dbfs total). for an input noise level of -75dbfs, this allows the snr to be decreased in steps of 1/8db or less. the i and q pn codes are offset in time to decorrelate them. the pn code is selected and enabled in the test control register (f800h). the pn is added to the signal after the mix with the three sign bits aligned with the most significant three bits of the signal, so the maximum level is - 12dbfs and the minimum, non-zero level is -108dbfs. the pn code can be 2 15 -1, 2 23 -1 or 2 15 -1 * 2 23 -1. cic filter next, the signal is filtered by a cascaded integrator/comb (cic) filter. a cic filter is an efficient architecture for decimation filtering. the power or magnitude squared frequency response of the cic filter is given by: where m = number of delays (1 for the HSP50216) n = number of stages and r = decimation factor. the passband frequency response for 1 st (n=1) though 5 th (n=5) order cic filters is plo tted in figure 8. the frequency axis is normalized to f s /r, making f s /r = 1 the cic output sample rate. figure 10 shows the frequency response for a 5 th order filter but extends the frequency axis to f s /r = 3 (3 times the cic output sample rate) to show alias rejection for the out of band signals. figure 9 uses information from figure 10 to provide the amplit ude of the first (strongest) alias as a function of the signal frequency or bandwidth from dc. for example, with a 5 th order cic and f s /r = 0.125 (signal frequency is 1/8 the cic output rate) figure 9 shows a first alias level of about -87 db. figure 9 is also listed in table form in table 47. the cic filter order is prog rammable from 0 to 5. the minimum decimation is 4. if the order is set to 0, there must be at least 4 clocks between samples or the decimation counter must be set to 4 to chose every 4th sample. the integrator bit widths are 69, 62, 53, 44, and 34 for the 1 st through 5 th stages, respectively, while the comb bit widths are all 32. the integrators are sized for decimation factors of up to 512 with 5 stages, 2048 with 4 stages, 32768 with 3 stages, and 65536 with 1 or 2 stages. higher decimations in the cic should be avoided as they will cause integrator overflow. in the HSP50216, the integrators are slightly oversized to reduce the quantization noise at each stage. pn value 2 -3 2 -4............ 2 -17 2 -18 sss x x xxxxxxxxxxxx x x pf () mf () sin f r ---- - ?? ?? sin ----------------------- - ?? ?? ?? ?? ?? 2n = HSP50216
10 backend data routing a cic filter has a gain of r n , where r is the decimation factor and n is the number of stages. because the cic filter gain can become very large with decim ation, an attenuator is provided ahead of the cic to prev ent overflow. the 24 bits of sample data are placed on the low 24 bits of a 69 bit bus (width of the first cic in tegrator) for a gain of 2 -45 . a 32 bit barrel shifter then provides a gain of 2 0 to 2 31 inclusive before passing the data onto the cic. the overall gain in the pre-cic attenuator c an therefore be programmed to be any one of 32 values from 2 -45 to 2 -14 , inclusive (see iwa=*004, bits 18:14). this shift factor is adjusted to keep the total barrel shifter and cic filter between 0.5 and 1.0. the equation which should be used to compute the necessary shift factor is: shift factor = 45 - ceiling(log 2 (r n )). note: with a cic order of zero, the cic shifter does not have sufficient range to route more than 10 bits to the back end since the maximum gain is 2 -14 (the least significant 14 bits are lost). back end section one back-end processing section is provided per channel. each back end section consists of a filter compute engine, a fifo/timer for evenly spacing samples (important when implementing interpolation filt ers and resamplers), an agc and a cartesian-to-polar coor dinate conversion block. a block diagram showin g the major functiona l blocks and data routing is shown above. the data input to the back end section is through the filter compute engine. there are two other inputs to the filter compute engine, they are a data recirculation path for cascading filters and a magnitude and d /dt feedback path for am and fm filtering. there are seven outputs from each back end processing section. these are i and q directly out of the filter compute engine (i2, q2), i and q passed through the fifo an d agc multipliers (i1, q1), magnitude (mag), phase (or d /dt), and the agc gain control value (gain). the i2 /q2 outputs are used when cascading back end stages. the routing of signals within the back end processing section is controlled by the filter compute engine. the routing information is embedded in the instruction bit fields used to define the digital filter being implemented in the f ilter compute engine. mag: i dphi/dt: q agc loop filter filter compute engine mux fifo/ timer agc mult cart to polar shift d/dt m u x x1, x2 x4, x8 m u x from cic path 0 path 1 path 2 (4:0) ext agc gain i1 q1 gain mag phase i2 q2 destination bit map (bits 28:18 of fir instructions bit field) 28 27 25 24 23 22 21 20 19 18 28 27 26, 25 24 23 22:18 agc loop gain select (path 01 only) update agc loop (path 01 only) path 00 - - immediate filter processor feedback path 01 - - fifo/agc path 10 - - direct out/cascade path 11 - - both 00 and 10 paths (for test) strobe output section (start se rial output with this sample) feed mag/phase back to filter processor filter processor sequence step number 26 HSP50216
11 filter compute engine the filter compute engine is a dual multiply-accumulator (mac) data path with a microc oded fir sequencer. the filter compute engine can implement a single fir or a set of filters. for example, the filter chain could include two halfband filters, a shaping (matc hed) filter and a resampling filter, all with different decima tions. the following filter types are currently supported by th e architecture and microcode: ? even symmetric with even # of taps decimation filters ? even symmetric with odd # of taps decimation filters (including hbfs) ? odd symmetric with even # of taps decimation filters ? odd symmetric with odd # of taps decimation filters ? asymmetric decimation filters ? complex filters ? interpolation filters (up to interpolate by 4) ? interpolation halfband filters ? resampling filters (under resampler nco control) ? fixed resampling ratio filter (within the available number of coefficients) ? quadrature to real filtering (w/ fs/4 up conversion) the input to the filter compute engine comes from one of three sources - a cic filter outpu t (which can also be another backend section), the output of the filter compute engine (fed back to the input) or the magnitude and d /dt fed back from the cartesian-to-polar coordinate converter. the number and size of the filters in the chain is limited by the number of clock cycles avai lable (determi ned by the decimation) and by the data and coefficient ram/rom resources. the data ram is 384 words (i/q pairs) deep. the data addressing is modulo in power-of-2 blocks, so the maximum filter size is 256. the block size and the block starting memory address for each filter is programmable so that the available memory can be used efficiently. the coefficient ram is 192 words deep. it is half the size of the data memory because filter coefficients ar e typically symmetric. roms are provided with halfband filter coefficients, resampling filter coefficients, and constants. the filter compute engine exploits symmetry where possible so t hat each mac can compute two filter taps per clock, by doing a pre-add before multiplying. in the case of halfband filters, t he zero-valued coefficients are skipped for extra efficiency. there is an overhead of one clock cycle per input sample for each filter in the chain (for writing the data into the data ram) and (except in special cases) a two clock cycle overhead for the enti re chain for program flow control instructions. the output of the filter comput e engine is routed through a fifo in the main output path. the fifo is provided to more evenly space the fir outputs when they are produced in bursts (as when computing resampling or interpolation filters). the fifo is four samples deep. the fi fo is loaded by the output of the filter when that path is sele cted. it is unloaded by a counter. the spacing of the output sample s is specified in clock periods. the spacing can be set from 1 (fall through) to 4096 samples m u x i q ram 384 words i q ramr/wb addra (8:0) addrb (8:0) s w a p s w a p a l u a l u r/d /dt 0..-23 inmux (1:0) 0..-23 a b 1..-23 1..-25 with rnd a b ramaen ramben iqswap ifunct qfunct coef s h f t r e g s h f t r e g l i m i t l i m i t r e g r e g r e g m u x m u x enfb, rndsel (2:0) shift (1:0) regen4 enlimit enhr1 enhr2 outsel down shift 0, 1, 2 places 9..-31 0..-23 0..-21 coef (21:0), shift (1:0) note: pipeline delays omitted for clarity iq HSP50216
12 (approximately the spacing fo r a 16ksps output sample rate when using 65msps clock) us ing iwa = *00ah bits 11:0. the number and order of the filteri ng in the filter chain is defined by a fir control program. th e fir control program is a sequence of up to 32 instruction words. each instruction word can be a filter or program flow in struction. the fi lter instruction defines a fir in the chain, specif ying the type of fir, number of taps, decimation, memory allocation, etc. for program flow, a wait for input sample(s) instruct ion, a loop counter load, and several jumps (conditional and unconditional) are provided. the HSP50216 evaluation board includes software for automatically generating fir control programs for most filter requirements. examples of programs fir cont rol programs are given below. the simplest filter program computes a single filter. it has three instructions (see sample filter #1program instructions below): the parameters of the fir (including type, number of taps, decimation and memory usage) are specified in the bit fields of the step 1 instruction word. to change the filtering the only other change needed is the number of samples in the wait threshold register (iwa = *00c, bits 9:0). the filter in this example requires 52 clock cycles to compute, allocated as follows: using a 65msps clock, the output sample rate could be as high as 65msps / 52 clocks = 1.25msps. the input sample rate to the fir from the cic filter would be 2.5msps. the impulse response length would be 38 sec (95 taps at 0.4 s/tap). each additional filter added to the signal processing chain requires one instruction step. as an example of this, a typical filter chain might consist of two decimate-by-2 halfband filters being followed by a shaping filter with the final filter being a resampling filter. the program for this case might be (see sample filter program #2 instructions below): sample filter #1 program step instruction 0 wait for enough input samples (equal to the decimation factor) 1fir type = even symmetric 95 taps decimate by 2 compute one output decrement wait counter memory block size 128 memory block start at 64, coefficient block start at 64 step size 1 output to agc 2 jump, unconditional, to step 0 sample filter #1 clock cycles calculation clock cycles function performed 48 clocks for fir computation (two taps/clock due to symmetry) 2 clocks for writing the input data into the data rams (decimate by 2 requires 2 inputs per output) 2 clocks for the program flow instructions (wait and jump) 52 total sample filter #2 program step instruction 0 wait for enough input samples (usually equal to the total decimation -- 8 in this case) 1fir type = even symmetry 15 taps halfband decimate by 2 compute four outputs memory block size 32 memory block start at 0 coefficient block start at 13 output to step 2 decrement wait count 2fir type = even symmetry 23 taps halfband decimate by 2 compute two outputs memory block size 32 memory block start at 32 coefficient block start at 24 output to step 3 3fir type = even symmetry 95 taps decimate by 2 compute one output memory block size 128 memory block start at 64 coefficient block start at 64 step size 1 output to step 4 4fir type = resampler increment nco 6 taps compute one output memory block size 8 memory block starts at 192 coefficient block start at 512 step size 32 output to agc 5 jump, unconditional, to 0 HSP50216
13 sample filter #2 requires: ? 32 + 32 + 128 + 8 = 200 data ram locations ? (95+1)/2=48 coefficient ram location (resampler and hbf coefficients are in rom). the number of clock cycles requ ired to comput e an output for sample filter #2 is calculated as follows: total decimation is 8, so the input sample rate for the fir chain (cic output rate) could be up to: f clk /(ceil(105/8)) = f clk /14. with a 65mhz clock, this would support a maximum input sample rate to the fir processor of 4.6mhz and an output sample rate up to 0.580mhz. the shaping filter impulse response length would be: (95 x 2)/580,000 = 82 s. the maximum output sample rate is dependent on the length and number of firs and their decimation factors. illustrating this concept with filter example #3, a higher speed filter chain might be comprised of one 19 tap decimate-by-2 halfband filter followed by a 30 tap shaping fir filter with no decimation. the program for this example could be: the number of clock cycles required to compute an output for sample filter #3 is calculated as follows: for filter example #3 and a 65msps input, the maximum fir input rate would be 65msps / ceil(26 / 2) = 5msps giving a decimate-by-2 output sample rate of 2.5msps. at 70msps, the fir could have up to 34 taps with the same output rate. channels 0, 1, 2 and 3 can be combined in a polyphase structure for increased bandwidth or improved filtering. filter example #4 will be used to demonstrate this capability. symbol rate of 4.096 msym. the desired output sample rate is 8.192msps. arran ge the four back end sections as four filters operating on the same cic output at a rate of 65.536mhz/4=16.384mhz, where t he factor of 4 is the cic decimation we have chosen. each channel computes the same sequence, offset by one output sample from the previous sample (see iwa = *00bh). each channel decimates down to 2.048m and then the sample filter #2 clock cycles calculation clock cycles function performed 20 halfband 1 compute clocks (5 per compute x 4 computes) 8 halfband 1 input sample writes (8 input samples) 14 halfband 2 compute clocks (7 per compute x 2 computes) 4 halfband 2 input sample writes (4 input samples) 48 95 tap symmetric fir, 2 clocks per tap 2 fir input sample writes (2 input samples) 6 resampler (6 taps, nonsymmetric) 1 resampler input sample write (1 input samples) 1 jump instruction 1 wait instruction 105 clock cycles per output sample filter #3 program step instruction 0 wait for enough input samples (2 in this case) 1fir type = even symmetry 19 taps halfband decimate by 2 compute one output memory block size 32 memory block start at 0 coefficient block start at 18 output to step 2 reset wait count 2fir type = even symmetry 30 taps decimate by 1 compute one output memory block size 64 memory block start at 32 coefficient block start at 64 step size 1 output to agc 3 jump, unconditional, to 0 sample filter #3 clock cycles calculation clock cycles function performed 6 19 tap halfband, one output 2 halfband input writes (2 input samples) 15 30 tap symmetric fir, 2 taps per clock 1 1 fir input write 11 wait 11 jump 26 clock cycles per output sample filter #3 program (continued) step instruction HSP50216
14 channels are multiplexed together in the output formatter to get the desired 8.1 92msps. the input sample rate to the final filter of each channel must meet nyquist requirements for the final output to assure that no information is lost due to aliasing. the number of fir taps availabl e for these requirements is calculated as follows: 65536/2048 = 32 clocks minus (8 writes + 1 wait + 1 jump = 10 clocks) = 22 clocks therefore, the number of taps available is: 22 x 2 = 44 taps. multiplexing the four outputs gives a final output sample rate of 8.192msps. the impulse response is 44 taps at 16.384m or 22 output samples (11 symbols at 4.096m). the agc loop filter output of channel 4 can be routed to control the forward agc gain c ontrol of all four channels. this assures that the gains of the four back end sections are the same. the gain error, however, is only computed from every fourth output sample. the back end processing sectio ns of two or more HSP50216s can be combined using the same polyphase approach, but the agc gain from one part cannot be shared with another part (except via the p interface), so polyphase filter using multiple parts would typically usually use a fixed gain. the filter sequencer is programmed via an instruction ram and several control registers. these are described below. instruction rams the filter compute engine is controlled by a simple sequencer supporting up to 32 steps. each step can be a filter or one of four sequence flow instructions - wait, jump (conditional or unconditional), load loop counter, or nop. there are 128 bits per instruction word with each word consisting of condition code selects, fir parameters and data routing controls. not all of the instruction word bits are used for all instruction types. the actual sequencer instruction is only 9 bits. the rest of the bits are used for filter parameters or for the loop counter preload. each sequence step is loaded in four 32-bit writes. the mapping of the bit fields for the instruction types is shown in the instruction bit field table that follows. these fir instruction words can be generated using software tool s provided with the HSP50216 evaluation board. when the filter is reset, the instruction pointer is set to 31 (the last instruction step). the read and write pointers are initialized on reset, so a reset must be done when the channel is initialized or restarted. a fixed offset can be added to the starting read address of one of the filters in the program. this function is provided to offset the data reads of the filt ers in a polyphase filter bank -- all filters in the bank will write the same data to the same ram location. to offset the computations the ram read address is offset. see iwa = *00bh for details. the instruction word bits (127:0) are assigned to memory words as follows: 31:0 to destination c c c c 0 0 0 1 0 x x x x x 0 0 63:32 to destination c c c c 0 0 0 1 0 x x x x x 0 1 95:64 to destination c c c c 0 0 0 1 0 x x x x x 1 0 127:96 to destination c c c c 0 0 0 1 0 x x x x x 1 1 where cccc is the channel number and xxxxx is the instruction sequence step number (0 - 31 decimal). note the phold bit in the filter compute engine control register (iwa = *00ah) must be set for the microprocessor to read from or write to the instruction or coefficient rams. sample filter #4 program step instruction 0 wait for enough input samples (8 in this case) 1fir type = even symmetry 44 taps decimate by 8 compute one output memory block size 64 memory block start at 0 coefficient block start at 64 step size 1 output to agc offset memory read pointers by 0, -2, -4, -6 2 jump, unconditional, to 0 HSP50216
15 filter sequencer ram addr gen b instruction ram, sequencer wait counter loop counter fir parameter fir# - write destination fir# - compute alias mask read pointer reg file write pointer reg file data path control rom ram addr gen a coef addr gen compute counters data address step size compute to compute fir type number of outputs taps/output reads/tap instr/tap start address ram addr initial offset ram addr offset step ram addr block to block step coef addr block start coef addr block size coef addr size per tap addr step size per output address offset resampler nco loop counter preload threshold decrement 1 decrement 2 new data, fir # reset sync fir output destination data path control signals data ram a read/write address data ram b read address enable coefficient offset read address ram addr block start ram addr block size ram addr step size 1 ram addr step size 2 ram addr block to block step ram HSP50216
16 instruction bit fields instruction bit fields bit positions function description 8:0 instruction instruction field bit mapping bit876543210 type wait0 0 xxxxccc fir 0 1 start incrrs decrsel decren ldlp decrlp enu/c jump1jjjjjccc (nops and loading the loop counter are special cases of the fir instruction). xxxx = ignored. jjjjj = jump destination (sequence step number). ccc = condition code. 000 = (waitcount threshold ) -- see iwa = *00ch, bits 9:0 for threshold details. 001 = waitcount threshold -- see iwa = *00ch, bits 9:0 for threshold details. 010 = loop counter 0. 011 = loop counter = 0. 100 = rsco tab (rsco - resampler nco carry output). 101 = rsco. 110 = sync (if enabled) or p controlled bit. 111 = always. start = load parameters and start filter computation, set to zero for no-ops, loop counter loads. incrrs = increment resampler during this filter. increments on start or at each fir output depending on pcontrol bit. decrsel = selects between two decrement values for the wait counter. decren = decrement wait count on starting this instruction. ldlp = load loop counter with the data in the i(20:9) bit field. the start bit should not be set when this bit is set. decrlp = decrement loop counter on starting this instruction. enu/c = enable u/c counter with this fir. this multiplies the data by 1, j, -1, -j. the multiplication factor changes each time the filter runs. 14:9 fir type fir parameter bit fields 14:9 fir type. 000000 nop. 000001 decimating fir, even symmetric, even # taps. 000010 decimating fir, even symmetric, odd # taps. 000011 decimating fir, odd symmetric, even # taps. 000100 decimating fir, odd symmetric, odd # taps. 000101 decimating fir, asymmetric. 001000 resampling fir, asymmetric. 001001 interpolating hbf. 100000 decimating fir, complex (asymmetric). notes: 1. regular interpolation firs are successive runs of a fir with no data address increment, but with coefficient start address increments. 2. decimating hbfs are even symmetric, odd number of taps but with different data step sizes. 3. u/c fir is a normal fir with the u/c bit enabled. 4. other codes may be added in the future. 17:15 steps per fir specifies the number of steps pe r fir instruction sequence (load with value minus 1) (set to 0 for all fir types exc ept complex which is set to 1). HSP50216
17 28:18 destination destination field bit mapping 28 27 26 25 24 23 22 21 20 19 18 agclfgn agclf path1 path0 os fb f4 f3 f2 f1 f0 agclfgnagc loop gain select. only applies to path 1. loop gain 0 or 1 if agclf bit is set. set to 0 (1 is a test mode for future chips). agclf agc loop filter enable. only applies to path 1. the agc loop is updated with the magnitude of this sample (path(1:0) = 01). path(1:0) back end data routing path selection. 00route output back to filter compute engine input to another fir in the filter chain. 01route output thru the fifo and agc forward path to the cartesian-to-polar coordinate converter conversion and output (i1, q1, magnitude, phase, gain) and also to route to a dis- criminator (i.e., d /dt fir). 10route output directly to the output, bypassing t he fifo and agc (i2, q2). this path also routes to next channel fir input. os enable output strobe. setting this bit generat es a data ready signal when the data reaches the output section and starts the serial output sequence (paths 1, 2, 3). if os is not set, there will be no output to the outside world from this channel, for that output calculation, but the data will be loaded into its output holding register (os would not be set when routing the data to another back end when cascading channels). fb feedback data path. when set, the magnitude and dp hi/dt from the cartesian-to-polar coor- dinate converter block are routed to the fi lter compute engine input (magnitude goes to the i input and dphi/dt goes to the q input). provided for discriminator filtering. f(4:0) filter select. for data recirculated to the input of the fir processor by path 0 or from the car- tesian to polar coordinate converter output, thes e bits tell which filter sequencer step gets it as an input. 31:29 round select 31:29 round select (add rounding bit at specified location). 000 2 -24 , use this code when downshifting is not used. 001 2 -23 010 2 -22 011 2 -21 100 2 -20 101 2 -19 110 2 -18 111 no rounding. provided for use with the c oefficient down-shift bits. 41:32 data memory block start memory block base address, 0-1023, 0-383 are valid for the HSP50216. 44:42 data memory block size 44:42 block size. 0 8 1 16 2 32 364 4 128 5 256 6 512 7 1024 (modulo addressing is used). 52:45 data memory block-to-block step 0-255, usually equal to the decimation factor for the fir in this instruction. instruction bit fields (continued) bit positions function description HSP50216
18 62:53 coefficient memory block start memory base address of coefficients, 0-1023, 0-511 are valid on the HSP50216. 63 reserved set to 0. 66:64 coefficient memory block size 66:64 memory block size 08 116 2 32 364 4 128 5 256 6 512 71024 (modulo addressing can be used, but is usually not needed. if not needed this bit field can always be set to 7). 75:67 number of fir outputs number of fir outputs (range is 1 to 512, load w/ desired value minus 1). this is usually equal to the total decimation that follows the filter. 84:76 read address pointer step read address pointer step (for next run). this is us ually equal to the filter decimation times the number of outputs from the instruction. 93:85 initial address offset initial address offset (to addrb). this is the offset from the start address to other end of filter. for symmetric filters, usually equal to -1 x (number of taps -1). 95:94 reserved set to 0 104:96 memory reads per fir output this is based on the number of taps (load with value below minus 1). type value symmetric, even number of taps(taps/2) or floor((taps+1)/2). symmetric, odd number of taps (taps+1)/2 or floor((taps+1)/2). decimating hbf(taps+5)/4. asymmetric taps. complex taps . resampling taps/phase (six taps per phase for the rom?d co efficients provided). interpolating hbf (taps+5)/4-1 . 106:105 clocks per memory read set to 0 for all but complex fir, which is set to 1. 115:107 data memory step size 1 (addra) step size for all but the last tap computation of the fir. set to -2 for hbf, -1 otherwise. 117:116 data memory step size 2 (addra) step size for last tap computation. set to -1. 117:116 step size 0 0 1 -1 2 -2 3 step size value. 119:118 data memory address offset step (addrb) step size for opposite end of symmetric filter. set to +2 for decimating hbf, to +1 for others (the b data is not used for asymmetric, resampling, and complex filters). instruction bit fields (continued) bit positions function description HSP50216
19 basic instruction set examples 1. wait for number of input samples > threshold 127:9 = 0 8:0 = 001 0000,0000,0000,0001h 2. jump unconditional 127:9 = 0 8:0 = 1jjjjj111b example: jump to st ep 0= 0000,0000,0000,0107h 3. jump rsco (jump on resampler nco carry output) 127:9 = 0 8:0 = 1jjjjj101b example: jump rsco, step 0= 0000,0000,0000,0105h 4. jump rsco (jump on no resampler nco carry output) 127:9 = 0 8:0 = 1jjjjj100b example: jump rsco , step 0 = 0000,0000,0000,0104h 5. nop single clock 127:9 = 0 8:0 = 010000000b nop1 = 0000,0000,0000,0080h 6. load loop counter 127:21 = 0 20:9 = loop counter prel oad (tested against 0) 8:0 = 010000100b example: ldlpcntr 14 = 0000,0000,0000,1c84h 122:120 coefficient memory step size (addrc) usually set to 1. 122:120 step size 00 1 1 2 2 34 4 8 5 16 6 32 764 125:123 coefficient memory block-to-block step (addrc) usually set to 0. 125:123 step size 0 0 1 1 2 2 3 4 4 8 5 16 6 32 7 64 127:126 reserved set to 0 instruction bit fields (continued) bit positions function description HSP50216
20 single fir basic program this is the basic program for a single fir. this program applies to decimation filters (including decx1) that are symmetric or asymmetric (but not complex). the fir output is routed through path a with the agc enabled. wait preload register this register (iwa register *00ch) holds the wait counter threshold and two wait counter decrement values. each is 10 bits. the wait counter counts filter input sa mples until the count is greater than or equal to the threshold. the wait counter then asserts a flag to the filter compute engine. the wait counter threshold is typically set to the total number of input samples needed to generate a filter output. a ?wait? instruction in the filter com pute engine waits for the wait counter flag signal before proceeding. the filter compute engine would then compute all the filters needed to produce an output and then would jump back to the ?wait? instruction. the wait counter is implemen ted with an accumulator. this allows the count to go beyond the threshold without losing the sample count. two bits in the fir instruction decrement the wait counter (subtract a va lue) and select the decrement value. the decrement value is typically the number of samples needed for an output (total decimation), though it can be a different value to ignore inputs and shift the timing. (the read pointer increment must be adjusted as well.) the filter compute engine sequencer does not count each input sample or track whether each filter is ready to run. instead, the wait counter is used to determine whether there are enough input samples to compute all the filters in the chain and get an output sample from the entire filter chain. this adds some additional delay since intermediate results are not precalculated, but it si mplifies the filter control. the number of samples needed is equal to the total decimation of the filter chain. for ex ample, with two decimate-by-2 halfband filters and a decimate-by-2 shaping fir, the total decimation would be 8 so 8 samples are needed to compute an output. hbf1 would compute four times to generate four inputs to hbf2. hbf2 would compute twice to generate the two samples that the shaping fir needs to compute an output. 0 - wait for enough samples 0000 0000 0000 0000 0000 0000 0000 0000 127:96 00000000h 0000 0000 0000 0000 0000 0000 0000 0000 95:64 00000000h 0000 0000 0000 0000 0000 0000 0000 0000 64:32 00000000h 0000 0000 0000 0000 0000 0000 0000 0001 31:0 00000001h 1 - fir 0000 0001 0101 1111 1111 100r rrrr rrrr 127:96 015ff---h 00tt tttt tttd dddd dddd 0000 0000 0111 95:64 -----007h 0000 1000 0000 0000 0000 1010 0000 0000 63:32 08000a00h 0000 1011 0000 0000 0fff fff0 1100 1000 31:0 0b00--c8h 2 - jump to step 0 0000 0000 0000 0000 0000 0000 0000 0000 127:96 00000000h 0000 0000 0000 0000 0000 0000 0000 0000 95:64 00000000h 0000 0000 0000 0000 0000 0000 0000 0000 64:32 00000000h 0000 0000 0000 0000 0000 0001 0000 0111 31:0 00000107h four bit fields must be filled in: f - filter type (this example applies to types 1-5) d - decimation (also loaded into wait threshold) t - number of taps minus 1 r - clocks/calculation (=floor((taps +1)/2) for symmetric, = taps for asymmetric) the rest of the instruction ram would ty pically be filled with nop instructions: 0000 0000 0000 0000 0000 0000 0000 0000 127:96 00000000h 0000 0000 0000 0000 0000 0000 0000 0000 95:64 00000000h 0000 0000 0000 0000 0000 0000 0000 0000 64:32 00000000h 0000 0000 0000 0000 0000 0000 1000 0000 31:0 00000080h HSP50216
21 resampler the resampler is an nco controlle d polyphase filter that allows the output sample rate to have a non-integer relationship to the input sample rate. the filter engine can be viewed conceptually as a fixed interpolate-by-32 filter , followed by an nco controlled decimator. the resampler nco is similar to the carrier nco phase accumulator but does not include the sin/cos section. it provides the resampler output pulse and associated phase information to logic that determines the nearest of the 32 available phase points for a given output sample. the center frequency (output sa mple rate) control is double buffered, i.e., the control word is written to one register via the microprocessor interface and then transferred to another (active) register on a write to the timing nco center frequency update strobe location (iwa register *009h) or on a synci (if enabled). as it is not possibl e to represent some frequencies exactly with an nco and therefore, phase error accumulates eventually causing a bit slip, the phase accumulator length has been sized to where the er ror is insignificant. at a resampler input rate of 1mhz, half an lsb of error in loading the 56-bit accumulator is 7*10 -12 degrees. after 1 year, the accumulated phase error is only 0.2*10 -3 of a bit (< 1/10 of a degree). the nco update by the filter compute engine is typically at the resampler's input rate, and is enabled by the incrrs bit in the filter instru ction word. the nco then rolls over at a fraction of the re sampler input rate. the output sample rate is (f in / 2 56 )*n, where f in is the resampler input rate and n is the phase accumulated per resampler input sample. n must be between 40000000000000h and ffffffffffffffh corresponding to decimations from 4 to (1 + 2 -56 ), respectively. generally, however, a range of 80000000000000h to ffffffffffffffh (providing decimation from 2 to (1 + 2 -56 ), respectively) is sufficient for most applications since integer decimation can be done more efficiently in the preceding cic and halfband filters. the resampler changes the sample rate by computing an output at each input which causes the nco to roll over. if an output is to be computed, the nearest of the 32 available points from the polyphase structure is used. because outputs are generated only on input samples which cause an nco roll over, output samples will in general not be evenly spaced. the fifo/timer block between the filter compute engine and the agc is provided to improve output sample spacing for presentation to the serial data output formatter section (see iwa=*00ah bits 11:0 description). if d/a converted directly, there would be artifacts from the uneven sample spacing, but if the samples are stored and reco nstructed at the proper rate (the nco rollover rate), the signal would have only the distortion produced by interpolation image leakage and the time quantization (phase jitter) due to the finite number of interpolation filter phases. the polyphase filter has 192 c oefficients implemented as 32 phases, each of which having 6 taps (6 x 32 = 192). these coefficients are provided in table 50. the stopband attenuation of the filter is gr eater than 60db, as shown in figures 13 - 15. the signal to total image power ratio is approximately 55db, due to the aliasing of the interpolation images. if the output is at l east 2x the baud rate, the 32 interpolation phases yield an effective sample rate of 64x the baud rate or approximately 1.5% (1/64 resampler input sample period) maximum timing error. agc the agc section provides gain to small signals, after the large signals and out-of-band noise have been filtered out, to ensure that small signals have sufficient bit resolution in the output formatter. the agc can also be used to manually set the gain. the agc optimizes the bit resolution for a variety of input amplitude signal leve ls. the agc loop automatically adds gain to bring small signals from the lower bits of the 24- bit programmable fir filter out put into the range of 20-bit and shorter words in the output section. without gain control, a signal at -72dbfs = 20log 10 (2 -12 ) at the input would have only 4 bits of resolution at the output if a 16 bit word length were to be used (12 bits less than the full scale 16 bits). the potential increase in the bit resolution due to processing gain of t he filters can be lost without the use of the agc. figure 1 shows the block diagram for the agc section. the fir filter data output is rout ed to the cartesian to polar coordinate converter after passing through the agc multipliers and shift register s. the magnitude output of the cartesian to polar coordinate converter is routed through the agc error detector, the agc error scaler and into the agc loop filter. this filtered error term is used to drive the agc multiplier and shifters, completing the agc control loop. the agc multiplier / shifter portion of the agc is identified in figure 1. the gain control from the agc loop filter is sampled when new data enters the multiplier / shifter. the limit detector detects overflow in the shifter or the multiplier and saturates the output of i and q data paths independently. the shifter has a gain from 0 to 90.31db in 6.021db steps, where 90.31db = 20log 10 (2 n ) when n = 15. the mantissa provides up to an additional 6.02db of gain. the gain in db from the mantissa is: 20log 10 [1+(x)2 -14 ], where x is the fractional part of the mantissa interpreted as an unsigned integer ranging from 0 to 2 14 - 1. thus, the agc multiplier / shif ter transfer function is expressed as: agc mult/shift gain = 2 n [1+ (x)2 -14 ] where n, the shifter exponent, has a range of 0HSP50216
22 in db, this can be expressed as: (agc mult/shift gain)db = 20 log 10 (2 n [1 + (x)2 -14 ]) the full agc range of the multiplier / shifter is from 0 db to 20log 10 [1+(2 14 -1)2 -14 ] + 20log 10 [2 15 ] = 96.329 db. the 16 bit resolution of the m antissa provides a theoretical am modulation level of -96dbc (depending on loop gain, settling mode and snr). this effectively eliminates am spurious caused by the agc resolution. the cartesian to polar coordinate converter accepts i and q data and generates magnitude and phase data. the magnitude output is determined by the equation: where the magnitude limits are determined by the maximum i and q signal levels into the cartesian to polar converter. taking fractional 2's complement representation, magnitude ranges from 0 to 2.329, where the maximum output is the agc loop feedback path consists of an error detector, error scaling, and an agc loop filter. the error detector subtracts the magnitude output of the coordinate converter from the programmable agc threshold value. the agc threshold value is set in iwa register *012h and is equal to 1.64676 times the desired magnitude of the i1/q1 output. note that the msb is always zero. the range of the agc threshold value is 0 to +3.9999. the agc error detector output has the identical range. shifter shifter exp=2 nnnn mantissa = limit det limiter limiter limit det 01.xxxxxxxxxxxxxx register 24 ifir qfir qagc iagc stt.ttttttttttttt unsigned ? threshold ? magnitude shift loop gain 1 man ? exp ? agcgnsel exp limit det upper limit ? lower limit ? p serial out msb = 0 agc loop filter agc multiplier/shifter 24 24 24 24 24 cartesian to polar coordinate converter 4 16 limiter 18 register 19 16 4 4 agc register 0 agc register 1 loop gain 0 man ? exp ? 16 28 agc error scaling agc msb = 0 error detector (range = 0 to 2.32887) (range = 0 to 1) (g = 1.64676) ? controlled via microprocessor interface. figure 1. agc functional block diagram (range = -2.18344 to 2.18344) + m p agc load en (11 mantissa 4 exponent) mantissa (s = 0) u x 16 r 1.64676 i 2 q 2 + = r 1.64676 1 2 1 2 + 1.64676x1.414 2.329 === HSP50216
23 the loop gain register values adjust the response / settling time of the agc loop. the loop gain is set in the agc error scaling circuitry, using four values in two sets of programmable mantissa and exponent pairs (see iwa register *010h). each set has both an attack and a decay gain. this allows asymmetric adjustment for applications such as vox systems where the signal turns on and off. in these applications, the gains would be set for fast attack and slow decay so that the part decreases the gain quickly when the signal turns on, but increases the gain slowly when the signal turns off (in anticipation of it turning back on shortly). for fixed gains, either set the upper and lower agc limits to the same value, or set the limits to minimum and maximum gains and set the agc attack and decay loop gains to zero. the mantissa, m, is a 4-bit valu e which weights the loop filter input from 0.0 to 15 / 2 4 = 0.9375. the exponent, e, defines a shift factor that provides additional weighting from 2 0 to 2 -15 . together the mantissa and exponent define the loop gain as given by, agc loop gain = m lg 2 -4 2 -(15-e lg ) where m lg is a 4-bit binary mantissa value ranging from 0 to 15, and e lg is a 4-bit binary exponent value ranging from 0 to 15. the composite (shifter and multiplier) agc scaling gain range is from 0.0000 to 2.329(0.9375)2 0 = 0.0000 to 2.18344. the scaled gain error can range (depending on threshold) from 0 to 2.18344, which maps to a ?gain change per sample? range of 0 to 3.275db / sample. the agc attack and decay gain mantissa and exponent values for loop gains 0 and 1 are programmed into iwa register *010h. the pdc provides for the storing of two values of agc attack and decay scaling gains to allow for quick adjustment of the loop gain by simply setting iwa register *013h bits 9 and 10 accordingly. possible applications include acquisition / tracking, no burst present / burst present, strong signal / weak signal, track / hold, or fast / slow agc values. the agc loop filter consists of an accumulator with a built in limiting function. the maximum and minimum agc gain limits are provided to keep the gain within a specified range and are programmed by 16-bit upper and lower limits using the following the equation: agc gain limit = (1 + m agc 2 -12 ) 2 e (agc gain limit)db = (6.02 )(eeee) + 20 log(1.0+0.mmmm mmmm mmmm) where m is a 12-bit mantissa value between 0 and 4095, and e is the 4-bit exponent ranging from 0 to 15. iwa register *011h bits 31:16 are used for programming the upper limit, while bits 15:0 are used to program the lower limit. the format for these limit values are: (31:16) or (15:0): e e e e m m m m m m m m m m m m for a gain of 0 1. m m m m m m m m m m m m * 2 e e e e and the possible range of agc limits from the previous equations is 0 to 96.328db. the bit weightings for the agc loop feedback elements are detailed in table 51. using agc loop gain, the agc range, and expected error detector output, the gain adju stments per output sample for the loop filter section of the digital agc can be given by agc slew rate = (1.5 db) (threshold - (mag * 1.64676)) x (m lg ) (2 -4 ) (2 -(15 - e lg ) ) the loop gain determines the growth rate of the sum in the loop accumulator which, in turn, determines how quickly the agc gain scales the output to the threshold value. since the log of the gain response is roughly linear, the loop response can be approximated by multiplying the maximum agc gain error by the loop gain. the expected range for the agc rate is ~ 0.000106 to 3.275db / output sample time for a threshold of 1/2 scale. for a fu ll scale error, the minimum non-zero agc slew rate would be approximately 0.0002db / output or 20db / sec at 100ksps. the maximum gain would be 6db / output. this much gain, however, would probably result in significant am on the output. the maximum agc response is given by: agc response max = (input)(cart/polar gain)(error det. gain)(agc loop gain)(agc output weighting) since the agc error is scaled to adjust the gain, the loop settles asympt otically to its final value. the loop settles to the mean of the signal. for example, if m lg = 0101 and e lg = 1100, the agc loop gain = 0.3125 * 2 -7 . the loop gain mantissas and exponents are set in iwa register *010h, with iwa register *013h selecting loop gain 0 or 1 and the settling mode. in the HSP50216, a synci signal will clear the agc loop filter accumulator if gwa register f802h bit 27 is set. the settling mode of the agc fo rces either the mean or the median of the signal magnitude error to zero, as selected by iwa register *013h bit 8. for mean mode, the gain error is scaled and used to adjust the gain up or down. this proportional scaling mode causes the agc to settle to the final gain value asymptotically. this agc settling mode is preferred in many applicati ons because the loop gain adjustments get smaller and smaller as the loop settles, reducing any am distortion caused by the agc. with this agc settling mode, the proportional gain error causes the loop to settle more slowly if the threshold is small. this is because the maximum value of the threshold minus the magnitude is smaller. also, the settling can be asymmetric, where the loop may settle faster for ?over range? signals than for ?under range? signals (or vice versa). in some applications, such as burst signals or tdma signals, a very fast settling time and/or a more predictable settling time is desired. the agc may be turned off or slowed down after an initial agc settling period. HSP50216
24 the median mode minimizes the settling time. this mode uses a fixed gain adjustment with only the direction of the adjustment controlled by the gain error. this makes the settling time independent of the signal level. for example, if the loop is set to adjust 0.5db per output sample, the loop gain can slew up or down by 16db in 16 symbol times, assuming a 2 samples per symbol output sample rate. this is called a median settling mode because the loop settles to where t here is an equal number of magnitude samples above and below the threshold. the disadvantage of this mode is that the loop will have a wander (dither) equal to the programmed step size. for this reason, it is advisable to set one loop gain for fast settling at the beginning of the burst and the second loop gain for small adjustments during tracking. in the median mode, the maximum gain step is approximately 3db / output. the step is fixed (it does not decrease as the error decreases) so a large gain will cause am on the output at least that large. the gain should be lowered after the settling. the fixed gain step is set by the programmable agc loop gain register iwa *010h. for median mode, the agc gain limits register sets the minimum and maximum limits on the agc gain. the total agc gain range is 96db, but only a portion of the range should be needed for most applications. for example, with a 16-bit output to a processor, the 16 bits may be sufficient for all but 24db of the total input range possible. the agc would only need to have a range of 24db. this allows faster settling and the agc would be at its maximum gain limit except when a high power signal was received. the agc may be disabled by setting both limits to the same value. the median settling mode is enabled by setting iwa register *013h bit 8 to 0 while the mean loop settling mode is selected by setting bit 8 to 1. cartesian to polar converter the cartesian to polar converter computes the magnitude and phase of the i/q vector. the i and q inputs are 24 bits. the converter phase output is 24 bits, msb?s routed to the output formatter and all 24 bits routed to the frequency discriminator. the 24-bit output phase can be interpreted either as two?s complement (-0.5 to approximately 0.5) or unsigned (0.0 to approximately 1.0), as shown in figure 2. the phase conversion gain is 1/2 . the phase resolution is 24 bits. the 24-bit magnitude is unsigned binary format with a range from 0 to 2.32. the magnitude conversion gain is 1.64676. the magnitude resolution is 24 bits. the msb is always zero. table 1 details the phase and magnitude weighting for the 16 bits output from the pdc. the magnitude and phase computation requires 17 clocks for full precision. at the end of the 17 clocks, the magnitude and phase are latched into a register to be held for the next stage, either the output formatter or frequency discriminator. if a new input sample arrives before the end of the 17 cycles, the results of the computatio ns up until that time, are latched. this latching means that an increase in speed causes only a decrease in resolution. table 2 details the exact resolution that can be obtained with a fixed number of clock cycles up to the required 17. the input magnitude and phase errors induced by normal snr values will almost always be worse than the cartesian to polar conversion. table 1. mag/phase bit weighting bit magnitude phase ( o ) 23 (msb) 2 2 180 22 2 1 90 21 2 0 45 20 2 -1 22.5 19 2 -2 11.25 18 2 -3 5.625 17 2 -4 2.8125 16 2 -5 1.40625 15 2 -6 0.703125 14 2 -7 0.3515625 13 2 -8 0.17578125 12 2 -9 0.087890625 11 2 -10 0.043945312 10 2 -11 0.021972656 92 -12 0.010986328 82 -13 0.005483164 72 -14 0.002741582 62 -15 0.001370791 52 -16 0.0006853955 42 -17 0.00034269775 32 -18 0.00017134887 22 -19 0.00008567444 12 -20 0.00004283722 0 (lsb) 2 -21 0.00002141861 0 + /2 - /2 0 /2 3 /2 000000 7fffff ffffff 3ff fff c00000 000000 800000 400000 bfffff i q i q ffffff 3fffff 400000 7fffff 800000 c00000 bfffff figure 2. phase bit mapping of coordinate converter output HSP50216
25 the enable signal for gating data into the coordinate converter is either the agc data ready signal or the resampler data ready signal. if the resampler is bypassed, the agc data ready signal is used and there is a delay of 6 clock cycles between the fir data being ready and the coordinate converter block sampling it. if the resampler is enabled, its data ready signal will be delaye d by 6 clocks (for the agc) plus the compute delay of the resampler block. this may cause the i/q to |r|/ output sample alignment to shift with the decimation. for this reason, it is recommended that the resampler/halfband filter block be bypassed when using this new data path. table 2. mag/phase accuracy vs clock cycles clocks magnitude error (% f s ) phase error (deg.) ? phase error (% f s ) 6 0.065 3.5 2 7 0.016 1.8 1 8 0.004 0.9 0.5 9 <0.004 0.45 0.25 10 <0.004 0.22 0.12 11 <0.004 0.11 0.062 12 <0.004 0.056 0.03 13 <0.004 0.028 0.016 14 <0.004 0.014 0.008 15 <0.004 0.007 0.004 16 <0.004 0.0035 0.002 17 <0.004 0.00175 0.001 ? assumes 180 o = f s . HSP50216
26 serial data output formatter section serial data output control register the serial data output control register contains sync position and polarity (synca, b, c or d), channel multiplexing, and scaling controls for the sd1x and sd2x (x = a, b, c or d) serial outputs (see microprocessor interface section, table 23, iwa *014h). channel routing mask the multiplexing mask bits for each channel (see microprocessor interface section table 23, iwa *014h bits 19:16 for sd1x or bits 15:12 for sd2x) can be used to enable that channel?s output to an y of the four serial outputs. these bits control the and gates that mask off the channels, so a zero disables the channel?s connection to that output. to configure more than one channel's output onto a serial data output, the sd1 serial outputs and syncs from each channel (0,1, 2 and 3) are brought to each of the sd1 serial output sections and the sd2 se rial outputs are brought to each of the sd2 serial output sections (the syncs are only associated with the sd1 serial outputs). there, the four outputs are and-ed with the multiplexing mask programmed in the serial data output control registers of channels 0 thru 3 and or-ed together. by gating off the channels that are not wanted and delaying the data from each desired channel appropriately, the channels can be multiplexed into a common serial output stream. it should be noted that in order to multiplex multiple channels onto a single serial data stream the channels to be mu ltiplexed must be synchronous. serial data output time slot content/format registers these four registers are used to program the content and format of the serial data output sequence time slots (see microprocessor interface section, tables 24 - 27, iwa *015h - *018h). there are seven data time slots that make up a serial data output stream. the number of data bits and data format of each slot is programmable as well as i1 m u x m u x o r & & & & & & & & & & & & fixed to float delay sync gen parallel to serial parallel to serial sequencer 1 sd2x 16 sequencer 2 round round m u x m u x r e g to p interface syncx sd1x output section zero q1 mag phase i2 q2 gain strobe zero note: each serial output has 7 time slots. each slot can contain i1, q1, i2, q2, mag, phase or d /dt. agc gain, or zeros. each slot can be 4, 6, 8, 10, 12, 16, 20, 24, or 32 (24 + 8 zeros) bits or disabled. ou tput 1 can also be 32-bit floati ng point. slots can be disabled. a disabled slot will be one clock wide if there are other active slot s following. a sync can be asserted with an y or all slots following. a sync can be ass erted with any or all slots in output 1. the serial output can be delayed from 0 to 4095 serial clock periods from the input strobe. the serial outputs are always msb first. the sync position applies to all time slots and can be one clock prior to the first data bit, aligned with the first data bit, or o ne clock after the last data bit. o r o r HSP50216
27 whether there will be a sync generated with the time slot (the syncs are only associat ed with the sd1 serial outputs). any of seven types of data or zeros can be chosen for each time slot. eight bits are used to specify the content and format of each slot. as an example, suppose we wanted to output 32-bit i and q values from channels 0 and 1 into the sd1a serial data output stream, we would program the following settings in the channel?s serial data output control and content/format registers: channel 0: delay = 0 (iwa = 0014h, bits 11:0 = 0); first data time slot = i, 32-bit, sync pulse generated (iwa = 0015h, bits 7:0 = 0xc9); second data time slot = q, 32-bit, no sync pulse (iwa = 0015h, bits 15:8 = 0x4a); third through seventh data time slot = zero and no sync, (iwa = 0015h, bits 31:16 = 0 and iwa = 0016h, bits 31:0 = 0); enable the sd1a serial output for this channel in the serial routing mask (iwa = 0014h, bit 16 = 1). channel 1: delay = 64 (iwa = 1014h, bits 11:0 = 0x40); first data time slot = i, 32-bit, sync pulse generated (iwa = 1015h, bits 7:0 = 0xc9); second data time slot = q, 32-bit, no sync pulse (iwa = 1015h, bits 15:8 = 0x4a); third through seventh data time slot = zero and no sync, (iwa = 1015h, bits 31:16 = 0 and iwa = 1016h, bits 31:0 = 0); enable the sd1a serial output for this channel in the serial routing mask (iwa = 1014h, bit 16 = 1). the resulting order is ch0 i first, then ch0 q, ch1 i, and ch1 q with sync pulses generated in the i data slots. the position of the sync pulses relative to the data slot may be programmed with iwa register *014h bits 25:24. setting delay = 64 offsets channel 1?s 32 bit i and q data by 64 clocks so that it immediatel y follows the 64 bits of data from channel 0. in this way channel 1?s first and second time slots follow channel 0?s second time slot. instead of using the delay to offset channel 1?s data, channel 0 could have been configured to output 32 bits of i in the fist slot, 32 bits of q in the second slot, 32 bits of zeros in the third slot and 32 bits of zeros in the fourth slot. channel 1 could then be configured to output 32 bits of zeros in the first and second slots, 32 bits of i in the third slot and 32 bits of q in the fourth slot. as the cha nnel outputs are or?d together, the zero slots do not in terfere with data slots. the HSP50216 microprocessor ( p) interface consists of a 16-bit bidirectional data bus, p(15:0), three address pins, add(2:0), a write strobe (wr ), a read strobe (rd ) and a chip enable (ce ). indirect addressing is used for control and configuration of the h sp50216. the control and configuration data to be loaded is first written to a 32-bit holding register at direct (e xternal) addresses add(2:0) = 0 and 1, 16 bits at a time. the data is then transferred to the target register, synchronous to the clock, by writing the indirect (internal) address of the target register to direct (external) address 2, add(2:0) = 2. the interface generates a synchronous one clock cycle wide strobe to transfer the data contained in the holding regi ster to the target register. the synchronization and write process requires 4 clock periods. new data should not be written to the holding register until after the synchronization period is over. HSP50216
28 microprocessor interface data reads can be direct, indi rect or fifo-like depending on the data that is being read. the status register is read directly at direct (exter nal) address 3, add(2:0) = 3. readback of internal registers and memories is indirect. the 16-bit indirect (internal) addre ss of the desired read source is first written to direct (exter nal) address 3, add(2:0) = 3, to select the data. the data ca n then be read at direct (external) addresses add(2:0) = 0 and 1 (bits 15:0 at address 0 and 31:16 at address 1). the data types available via the indirect read are listed in the tables of indirect read address (ira) registers. (note that the phold bit contained in the target register at indirect write address (iwa) = *00ah must be set to susp end the filter comput e engine before the coefficient ram and instruction bit fields can be written to or read from.) the HSP50216 output data from the four channels is available through the microprocessor interface as well as from the serial data outputs. a fifo-like interface is used to read the output data through the microprocessor interface. when new output data is availabl e, it is loaded into a fifo in a user programmed order (for details on the programming order see global write address (gwa) = f820h - f83fh). it can then be read, 16 bits at a time, at direct address 2, add(2:0) = 2. at the end of each read, the fifo counter is advanced to the next location. this allows a dma controller to read all of the data with succe ssive reads to a single direct address. no writes or other in teraction is required. the fifo counter is reset and reloaded by each interrupt signal, see gwa f802h. new data in the fifo is also indicated in the status register located at di rect address add(2:0) = 3 if a polled mode is preferred. the eight data types available, for each of the four channels, vi a this interface are: i(23:8), i(7:0)+8 zeroes, q(23:8), q( 7:0)+8 zeroes, mag(23:8), mag(7:0)+8 zeroes, phase (1 5:0), and agc (15:0). the upper bits of i, i.e., i(23:8), and q, i.e., q(23:8), are not rounded to 16 bits. this interface can read the data from all the channels that are synchr onized. however, because a common fifo is used and the fifo is reset and reloaded by each interrupt, it cannot be used for asynchronous channels. the direct address map for the microprocessor interface is shown in the table of microprocessor direct read/write addresses and the procedures for reading and writing to this interface are provided below. the bit field details for each indirect read and write address is provided in the table of indirect read address (ira) r egisters, tables of indirect write address (iwa) registers (tables 3 - 34) and tables of global write address (gwa) registers (tables 35 - 45). mux 3 2 1 0 r e g > r e g > f f > f f > f f > f f > r e g > and m u x e s d e c o d e rd p(15:0) wr a(2:0) clk = 0 = 1 = 2 or 3 = 2 en en en 15:0 31:0 31:16 15:0 31:16 31:0 internal read data bus internal read signal sync?d wr to target registers internal address bus internal write data bus special low metastability cell rst ce (gating not shown) l a t c h from output fifo status g a t i n g HSP50216
29 p read/write procedures to write to the internal registers: 1. load the indirect write holding registers at direct address add(2:0) = 0 and 1 with the data for the internal register (16 or 32 bits depending on the internal register being addressed). 2. write the indirect write addr ess of the internal register being addressed to direct address add(2:0) = 2 (note: a write strobe to transfer the co ntents of the indirect write holding register into the tar get register specified by the indirect address will be generated internally). 3. wait 4 clock cycles before performing the next write to the indirect write holding registers. to write to the internal instruction/coefficient rams: 1. put the filter compute engine of the desired channel into the hold mode by setting bit 31 of the filter compute engine / resampler control register located at iwa = *00ah (note: the * is equal to 0, 1, 2 or 3 depending on the channel being addressed). by setting bit 31 all fir processing for the channel addressed will be stopped. 2. load the indirect write holding registers at direct address add(2:0) = 0 and 1 with the data for the internal ram location. 3. write the indirect write ad dress of the internal ram location being addressed to direct address add(2:0) = 2 (note: a write strobe to transfer the contents of the indirect write holding register into the ram location specified by the indirect address will be generated internally). 4. wait 4 clock cycles before performing the next write to the indirect write holding registers. 5. after all data has been loaded, set the phold bit back low. to read internal registers: 1. write the indirect read addr ess of the internal register being addressed to direct address add(2:0) = 3. 2. perform a read of the indirect read holding registers at direct address add(2:0) = 0 and 1. to read data outputs: 1. set up the p fifo read order control register (located at global write address (gwa) = f820h - f83fh). 2. wait for interrupt or check flag. 3. data can then be read, 16 bits at a time, at direct address 2, add(2:0) = 2. 4. repeat step 3 for desired number of words. 5. go to step 2. to read instruction/coefficient values: 1. put the filter compute engine of the desired channel into the hold mode by setting bit 31 of the filter compute engine / resampler control register located at iwa = *00ah (note: the * is equal to 0, 1, 2 or 3 depending on the channel being addressed). 2. write the indirect read address (ira) of the internal ram/rom location being addressed to direct address add(2:0) = 3. 3. wait 4 clock cycles. 4. read the data at direct address add(2:0) = 0 and 1. 5. after all the data has been read, set the phold bit back low. recommended HSP50216 configuration procedure following a hardware reset (i.e. resetb is pulsed low): 1. load global write address registers gwa f800 - gwa f808 and gwa f820 - gwa f83f. 2. for each signal processing channel (0-3): a. set mphold bit located at indirect write address register iwa *00a - 31. b. load filter compute engine instruction rams. c. load filter compute engine coefficient rams. d. load iwa registers *000 - * 019. (clear the mphold bit in register iwa *00a - 31). e. wait 32 clocks (clk) for th e reset to complete in the filter compute engine. 3. generate a synci to enable the input data or to synchronize the processing to external events or generate a synco by writing to gwa f809. note: for the latter method, the synco pin must be connected to the synci pin. recommended HSP50216 channel reconfiguration procedure: 1. disable the serial output for the desired channel in register gwa f801 - 3, 2, 1 or 0. 2. disable the interrupts from the channel in register gwa f802 - 31, 23, 15, or 7. 3. set the mphold bit in register iwa *00a - 31 to give the processor access to the filter compute engine instruction rams and coefficient rams. 4. load the new filter configuration. 5. load any other channel registers. 6. clear the mphold bit in register iwa *00a - 31. 7. do a software channel reset by writing to iwa *019. 8. enable the serial outputs (gwa f801) and interrupts (gwa f802). 9. generate a synci to enable the input data or to synchronize the processing to external events or generate a synco by writing to gwa f809. note: for the latter method, the synco pin must be connected to the synci pin. HSP50216
30 table of microprocessor direct read/write addresses add(2:0) pins register description 0 wr indirect write holding register, bits 15:0. 1 wr indirect write holding register, bits 31:16. 2 wr indirect write address register for internal target regi ster (generates a write strobe to transfer contents of the write holding register into the target register specified by the indirect address, s ee also tables of indirect address registers). 3 wr indirect read address register (used to select the read source of data - uses the same register as direct address 2 but generates a read strobe (for rams and agc) as needed instead of a write strobe). 0 rd indirect read, bits 15:0. 1 rd indirect read, bits 31:0f. 2 rd read register (fifo) - reads fifo data from output section (this location reads output data in the order loaded in global control indirect address registers f820- f83f. the fifo is automatically incremented to the next data location at the end of each read). 3 rd status register p(15:0) bit description 15:12 unused. 11:6 read non-bus input pins (enix , reset , synci). 11 reset (note: this bit is inverted with respect to the reset input pin). 10 enia . 9enib . 8enic . 7 enid . 6 synci. 5:2 mask revision number. 1 level detector integration done. active high. 0 new fifo output data available (used for polling mode vs interrupt mode) active low. HSP50216
31 tables of indirect write address (iwa) registers note: these indirect write addres ses are repeated for each channel. in the addresses below, the * field is the channel select nibble. these bits of the indirect address select the target channel register for the data. values of 0 through 3 and f are valid. a channel selec t nibble value of f is a special case which writes the data to the same location in each of the fo ur channels simultaneously. table 3. channel input select/format register (iwa = *000h) p(15:0) function 15:13 channel input source selection - selects as the data input fo r the channel specified in the i ndirect address either a(15:0 ), b(15:0), c(15:0), d(15:0) or the p test input register as shown below: 15:13 source selected 000 a(15:0) 001 b(15:0) 010 c(15:0) 011 d(15:0) 100 p test input register. this is provided for testing and to ze ro the input data bus when a channel is not in use. the glo- bal write address register for the p test input register is f807h. 12 p test register input enable selection: 1 bit 11 of this register is used as the input enable. 0 a one clock wide pulse generated on each write to lgwa f808h is used as the input enable. select 0 to write test data into the part. select 1 to input a constant or to disable the input for minimum power dissipation when an nco/mixer/cic section is unused. 11 p input enable. when bit 12 is set, th is bit is the input enable for the p test register input. active low: 0 enabled 1 disabled. 10 parallel data input format: 0 two?s complement (-full scale = 1000...0000, zero = 0000...0000, +full scale = 0111...1111). 1 offset binary (-full scale = 0000...0000, zero = 1000...0000, +full scale = 1111...1111). 9 fixed/floating point: 0 fixed point. 1 floating point. the 16-bit input bus is divided into mantissa and exponent bits grouped either 13/3 or 14/2 depending on bits 8 and 7. see text. 8:7 floating point mantissa size select. the 16-bit data input is grouped as a 13/3 or 14/2 mantissa/exponent word. these contro l bits select the mantissa/exponent grouping, add an offset to the exponent and set the shift control saturation level: 00 11/3: bits 15:5 are mantissa, 2:0 are exponent. 01 12/3: bits 15:4 are mantissa, 2:0 are exponent. 10 13/3: bits 15:3 are mantissa, 2:0 are exponent. 11 14/2: bits 15:2 are mantissa, 1:0 are exponent. see the exponent tables contained in the input select/format block section. 6:4 de-multiplex control. these control bits are provided to select a channel from a group of multiplexed channels. up to 8 mult iplexed data streams can be demultiplexed. these control bits select how many clocks after the enix signal to wait before taking the input sample. enix should be asserted for one clock period and aligned with the first channel of the mu ltiplexed data set. for example, if four streams are multiplexed at half the clock rate, enix would align with the first clock peri od of the first stream, the second would start two clocks later, the next 4 clocks after enix , etc. the samples are aligned with enix (zero delay) at the input of the nco/mixer/cic stage at the next enix . 000 zero delay 111 7 clock periods of delay. all values from 0 through 7 are valid. 3 interpolated/gated mode select: 0 gated. the carrier nco and cic are updated once per clock when enix is asserted. 1 interpolated. the cic is updated every clock. the carrier nco is updated once per clock when enix is asserted. the input is zeroed when enix is high. HSP50216
32 2 enable cof/cofsync inputs. when set, this bi t enables two bits from the d(15:0) input data bus to be used as a carrier offset frequency input. 1 enable sof/sofsync inputs. when set, this bit enables two bits from the d(15:0) input data bus to be used as a resampler offse t frequency input. 0 enable pn. when set, a pn code, weighted by the gain in location *001, is added to the input samples at the output of the mixe r. table 3. channel input select/format regi ster (iwa = *000h) (continued) p(15:0) function table 4. pn gain register (iwa = *001h) p(31:0) function 31:16 reserved, set to all 0?s. 15:0 pn generator gain register. this input is provided to reduce the sensitivity of t he receiver. a pn code, weighted by the va lue in this location, is added to the data at the output of the mixer. addi ng noise has the effect of increasing the receiver noise figure. one reason to do this would be to decrease the basesta tion cell size in small st eps. this method is very accurate and repeatable and can b e done on a fdm channel by channel basis. it does , however, reduce the overall dynamic range. an alternate way is to add attenuat ion at the rf and adjust the whole range upward. this does not redu ce the overall range but only shift it, with the shift being don e on all channels simultaneously. table 5. cic decimation factor register (iwa = *002h) p(15:0) function 15:0 load with the desired cic decimation factor minus 1. table 6. cic destination fir and output enable/disable register (iwa = *003h) p(15:0) function 15:6 set to zero. 5:1 cic output destination (fir # in fir processor). usually set to 00001. 0 cic output enable. active high. when low, the data writes from the cic to the filter compute engine are inhibited. table 7. carrier nco/cic control register (iwa = *004h) p(31:0) function 31:19 reserved, set to zero. 18:14 cic barrel shift control. 00000 is the minimum shift factor and 11111 is maximum shift fa ctor. this compensates for the cic filter gain of r n , where n is the number of enabled cic stages and r is the cic decimation factor . the equation used to compute the shift factor is: shift factor = 45 - ceiling(log 2 (r n )). examples: nrshift factor 5 512 0 5830 13:9 cic stage bypasses. the integrator/comb pairs are numbered 1 th ru 5 with 1 being the first integrator and first comb. bit 1 3 bypasses the first integrator/comb pair, bit 12 bypasses the second, etc. the first integrator is the largest. typically, the stages are enabled starting with stage 1 for maximum decimation range. 8:6 carrier phase shift. phase shifts of n*( /4), n = 0 to 7. 5 clear feedback (test signal or for mixer bypass). 4 nco clear feedback on load. 3 update frequency on synci. redundant. set to1. see gwa register f802h. HSP50216
33 2:1 number of carrier offset frequency (cof) serial input bi ts. the format is 2?s comple ment, early sync, msb first: 00 8 01 16 10 24 11 32 0 enable serial carrier offset frequency (zeros the data already lo aded via the cof/cofsync pins). to disable the cof shifting s ee iwa register *000h. table 7. carrier nco/cic control register (iwa = *004h) (continued) p(31:0) function table 8. carrier nco center frequency register (iwa = *005h) p(31:0) function 31:0 carrier center frequency (ccf): this is the frequency control for the carrie r nco. the center frequency control is doubl e buffered. the contents of this regist er are transferred to the active register on a write to the ccfstro be location or on a synci (if load on synci is enabled). the carrie r center frequency is: ccf*f clk /(2 32 ). ccf is a twos complement number and has a range of -2 31 to (2 31 -1). f clk is the input sample rate (enix assertion rate) for gated mode and the clock rate for interpolated mode. table 9. carrier nco center frequency update strobe register (iwa = *006h) p(15:0) function n/a writing to this address generates a strobe that transfers the ccf value to the active frequency register. the transfer to th e active register can also be done using the synci pin to synchronize the transfer in multiple parts or to synchronize to an external ev ent. the value in the active register can be read at this address (the center frequency control before the serially loaded offset va lue is added). to read the value, either write this address to a(1:0) = 11 and then read at a(1:0) = 00 and 01, or read the value at a (1:0) = 00 and 01 after writing to this address and before writing a new address to either a(1:0) = 10 or 11. table 10. timing nco frequency control register, msw (iwa = *007h) p(31:0) function 31:0 these are the upper 32 bits of the 56-bit ti ming (resampler) nco center frequency control. table 11. timing nco frequency control register, lsw (iwa = *008h) p(31:0) function 31:8 these are the lower 24 bits of the 56-bit timing (resampler) nco center frequency control. 7:0 unused, set to zero. table 12. timing nco center frequency load strobe register (iwa = *009h) p(31:0) function n/a for wr 31:0 for rd a write to this location will update the resampler nco center fr equency. the upper 32 bits of the active register can be read a t this address. table 13. filter compute engine/resampler control register (iwa = *00ah) p(31:0) function 31 phold. when set, this bit stops the filter compute engine and allows the p access to the instruction and coefficient rams for reading and writing. on the high to low transition, the filter compute engine is reset (the read and write pointers are reset a nd the instruction at location 31 is fetched). 30 pshiftzerob. this bit, when set to zero, dis ables the coefficient shift bits (bits 9:8 of the master register when coefficient loading). 29 pen l imit. this bit disables the data path satura tion logic. provided for test. active high. set to 0 to disable the normal rom controlled limiting (anded with normal signal). HSP50216
34 28:24 pz(4:0). these bits, when set to zero, zero the corresponding read pointer address bits. this al lows the pointers to be aliased , i.e., multiple filters can access and/or modify the same pointer. th ey are provided to change filters, coefficients or decimation ove r a sequence. 23 unused, set to 0. 22 timing (resampler) nco ensync. if this bit is se t, the center frequency is updated on a synci. set to 1. 21:20 rsrvrs(1:0). set to 01. 19 beginning/end . this bit selects whether the resampler nco is updated at the beginning of a fir computation or at the end of each fir output computation. usually, the resamp ler will be updated once at the beginning of each resampler computation and this wil l be bit set to 1. 1 once at the beginning of the fir instruction. 0 at the last tap of each of the instruction?s fir computations (once per output). 18 rsmodeselect. this bit selects whether the resa mpler is a phase shifter or a frequency shifter. 0 phase shift. it uses the top 5-bits of the timing nco frequency to determine a phase shift and disables feedback in the timing nco phase accumulator -- effect of the resampler is a constant phase shift. 1 frequency shift. effect of the resampler is a change in the sample rate. 17 rsco. this bit is provided to force the resampler nco carry when using the re sampler as a phase shifter rather than for a fre quency shift. this bit must be set for phase shifting and cleared for frequency shifting. (the bit is or -ed with the normal carry.) 16 rs nco clear phase accumulator feedback on load. when this bit is set, the feedback in the resampler nco phase accumulator is zeroed whenever the center frequency word is updated. this forces the nco to a known phase so the phase of multiple channels can be aligned. 15 force nco load. this bit, when set, zeroes the feedback in the re sampler nco phase accumulator. this is provided for test or to use the resampler for phase instead of frequency shifting. 14 enable rs freq offset. this bit, when set, enables the seria lly loaded resampler offset frequency word. when zero, the offset is zeroed. to disable the shi fting, see iwa register *000h. 13:12 serial input word size. these bits select the number of bits in the resampler offset frequency word (loaded serially via sof/sofsync). 00 8 bits 01 16 bits 10 24 bits 11 32 bits 11:0 fifodelay. a fifo is provided at the output of the filter compute engine to smooth the sample spacing when using the resamp ler or interpolation firs. in these filters, the outputs can be produced in bursts or with gaps. the fifo takes the samples in and out puts them based on a counter timeout. if the fifo is empty and the counter is at its terminal count (hold state), the data is passed through and the counter is reloaded. if the counter is not at terminal c ount, the data is held in the fifo until the counter times out. the fifo can hold up to 4 samples. the delay is programmed in clock peri ods. the value programmed is one less than the number of clocks of delay. set to 0 for a delay of one (fall through) . the delay should be programmed to slightly less th a n the desired spacing to prevent overflow. table 13. filter compute engine/resampler co ntrol register (iwa = *00ah) (continued) p(31:0) function table 14. filter start offset register (iwa = *00bh) p(15:0) function 13:9 ram instruction number to which the offset is appl ied. 0-31. aliasing applies. used for polyphase filters. 8:0 amount of offset. offsets the data ram address for filter #n. this is used to o ffset the channels from each other when break ing the processing up among multiple channels for poly phase filters. for example, four channels can receive the same data at 8 msps, fil ter and decimate by 8 to output at 1mhz. if the computations are offs et by 2 samples each, then the outputs of the four channels ca n be multiplexed together to get an output sample rate of 4msps. with a 64msps clock, the composite filter could have more than 100 taps where a single channel would only be capable of around 24 taps at a 4mhz output. except in very rare circumstances, th is value should be a negative number. HSP50216
35 table 15. wait threshold/decrement value register (iwa = *00ch) p(31:0) function 31 ptestbit. this bit is provided as a micr oprocessor controlled condition code for t he filter compute engine for conditional exec ution or synchronous startup. active high. 30 set to 0. 29:20 decrement value 1. positive number. 19:10 decrement value 0. positive number. usually set equal to the threshold (bits 9:0). 9:0 threshold. number of samples needed to run a filter set and produce an output. table 16. reset write pointer of fset register (iwa = *00dh) p(15:0) function 15:9 set to zero. 8:0 this parameter is the offset between filter compute engine r ead and write pointers on filter compute engine reset. on reset, the read and write pointers for all the filters are loaded, the read pointer with zero and the write pointer with this value. set to zer o for a single filter and two for a multi-filter chain. table 17. agc gain load register (iwa = *00eh) p(15:0) function 15:0 this location loads the agc accumulator. if the loop attack/dec ay gain is set to zero and this value is within the agc gain limits, the agc will hold this value. if not, the agc will be set to this gain (or to a limit) and then start to settle. format is 4 exponent bits (15:12), and 12 mantissa bits, (11:0). table 18. agc gain read strobe register (iwa = *00fh) p(15:0) function 15:0 for rd ; n/a for wr writing to this location will sample the agc loop filter output (forward gain value) to stabilize it for reading. the value is read from this location after waiting the 4 cl ocks required for read synchronization. table 19. agc loop attack/decay gain values register (iwa = *010h) p(31:0) function 31:24 loop gain 0, decay gain value (signal decay, incr ease gain) 31:28 = eeee (exponent), 27:24 = mmmm (mantissa). 23:16 loop gain 1, decay gain value 23:20 = eeee (exponent), 19:16 = mmmm (mantissa). 15:8 loop gain 0, attack gain value (signal arrival, decr ease gain) 15:12 = eeee (exponent), 11:8 = mmmm (mantissa). 7:0 loop gain 1, attack gain value 7:4 = eeee (exponent), 3:0 = mmmm (mantissa). table 20. agc gain limits register (iwa = *011h) p(31:0) function 31:16 upper gain limit. see agc section. 15:0 lower gain limit. see agc section. table 21. agc threshold register (iwa = *012h) p(31:0) function 16 enables dphi/dt update for non-fed back data. discriminator output is not filtered. 15:0 agc threshold. equals 1.64676 times th e desired magnitude of the i1/q1 output. HSP50216
36 table 22. agc/discriminator control register (iwa = *013h) p(15:0) function 15:11 set to zero. 10 p agc loop gain select. 9 enable filter compute engine control of ag c loop gain. when this bit is set, bit 28 in the filter compute engine destination field selects which loop gain to use with that filter output?s gain erro r. setting bit 10 overrides this bit and forces a loop gain 1. 10:9 function 00 loop gain 0 ( p controlled) 10 loop gain 1 ( p controlled) 01 loop gain controlled by filter compute engine 11 loop 1 ( p override of filter compute engine) 8 mean/median. this bit controls the settling mode of the agc. mean mode settles to the mean of the signal and settles asymptoti cally to the final value. median mode settles to the median and settles with a fixed step size. this mode settles faster and more pre dictably, but will have more am after settling. 1 mean mode 0 median mode 7 set this bit to 1 to get a dphi/dt output without having to feedback through the filter compute engine . 6 unused. set to zero. 5 phaseoutputsel 1d /dt 0 phase 4:3 discshift(1:0). shifts the phase up 0, 1, 2, or 3-bit positions, discarding the bits shifted off the top. this makes the pha se modulo 360, 180, 90, or 45 degrees to remove psk modulation. the resulting phase is 18 bits. 2:0 discdelay(2:0). sets the delay, in sample times, for the d /dt calculation. 000 1 111 8 table 23. serial data output control register (iwa = *014h) p(31:0) function 31:29 set to zero. 28 sync polarity 1 active low (low for one serial clock per word with a sync). 0 active high. 27:26 reserved, set to zero. 25:24 sync position. this applies to all time slots in the serial output. the sync progr amming is associated with the sd1x seria l output data stream (x = a, b, c, or d). 00 sync is asserted during the serial cl ock period prior to the first data bi t of the serial word (early sync). 01 sync is asserted during the cloc k period following the last data bit of the word (late sync). 1x sync is asserted during the serial clock period of t he first data bit of the seri al word (coincident sync). 23:22 reserved, set to zero. HSP50216
37 21:20 magnitude output scale factor. the magnitude output of the ca rtesian to polar coordinate c onversion has bits weighted as: 2 (2 1 0.-1 -2 -3 -4 . . . ) the gain in the conversion is 0.82338. when us ing 16 bits, the range is such that the lsb has a weight of 0.00007 and the maxim um output is 2.32, both after the conversion gai n. this corresponds to an i/q vector lengt h of -83dbfs to +3dbfs. these control bits add gain (with saturation) for more resolution at the bottom of the scale. a code of 00 passes the magnitude unchanged, 01 shif ts the magnitude up one bit position? 10 shifts by 2 positions and 11 shifts up three posit ions. the resulting bit weights and ran ge (after conversion gain) for the unsigned numbers are: code bit weights dbfs 00 2 1 0 -1 -2 . . . -11 -12 -13 +3 to -83 01 1 0 -1 -2 -3 . . . -12 -13 -14 +3 to -89 10 0 -1 -2 -3 -4 . . . -13 -14 -15 +1.7 to -95 11 -1 -2 -3 -4 -5 . . . -14 -15 -16 -4.3 to -101 the upper limits on codes 00 and 01 are the same, but 01 has no leading zero. 19:16 serial data output sd1 rout ing mask. 0 disables. 1 enables. bit enabled output 16 enables the serial output for this channel to pin sd1a. 17 enables the serial output for this channel to pin sd1b. 18 enables the serial output for this channel to pin sd1c. 19 enables the serial output for this channel to pin sd1d. 15:12 serial data output sd2 rout ing mask. 0 disables. 1 enables. bit enabled output. 12 enables the serial output for this channel to pin sd2a. 13 enables the serial output for this channel to pin sd2b. 14 enables the serial output for this channel to pin sd2c. 15 enables the serial output for this channel to pin sd2d. 11:0 output hold-off delay. this parameter adds additional delay from the output of the filter compute engine to start of the se rial output stream for multiplexing channels. load with the desired delay (0 = zero, 1 = one, 2 = two, etc.). table 23. serial data output control register (iwa = *014h) (continued) p(31:0) function table 24. serial data output 1 content/format register 1 (iwa = *015h) p(31:0) function 31:24 fourth serial slot in serial data output 1 (sd1x). x = a, b, c or d. see bits 7:0 for functional description of bits 31:24 . 23:16 third serial slot in serial data output 1 (sd1x). x = a, b, c or d. see bits 7:0 for f unctional description of bits 23:16. 15:8 second serial slot in serial data output 1 (sd1x). x = a, b, c or d. see bits 7:0 for f unctional description of bits 15:8. HSP50216
38 7:0 first serial slot in serial data output 1 (sd1x). x = a, b, c or d. bit function 7 sync generated. when set, a sync pulse is generated with the dat a slot (serial data output 1 only, i.e., the sync is only associated with output 1). set to zero for output 2, sd2x. 6:3 word width/format. all fixed point data is twos complement. the data is rounded (asymmetrical ly, with saturation) to the desired number of bits. 0000 0-bit, fixed point (actually 1-bit position is used). 0001 4-bit, fixed point. 0010 6-bit, fixed point. 0011 8-bit, fixed point. 0100 10-bit, fixed point. 0101 12-bit, fixed point. 0110 16-bit, fixed point. 0111 20-bit, fixed point. 1000 24-bit, fixed point . 1001 32-bit fixed (8 lsbs are zeroed). 1010 32-bit, floating point, ieee format. all other codes are invalid. note: floating point format is only available on the serial da ta output 1. code 1010 is invalid on serial data output 2. 2:0 data type 000 zeros 001 i1 (data routed from fifo and agc path). 010 q1 (data routed from fifo and agc path). 011 magnitude of i1/q1. 100 phase (or d /dt) of i1/q1. 101 i2 (data routed directly fr om the filter processor). 110 q2 (data routed directly from the filter processor). 111 agc gain of i1/q1 path. the filter processor must be programmed appropriately to route the data to i1/q1 or i2/q2. note: disable a slot by setting the 8-bit word to 00h. when disabled, a slot still uses one clock period. if, for example, the slots are programmed to 16-bit, disabled, 16-bit, there would a one clock idle period between the two 16-bit data words. if a new data sample occurs before the current set of data has been output, the new data will preempt the output and the first slot of the new data will begin immediately. if a late sync was programmed, it will not occur. 0123456789abcdef0123456789abcdef i, q 012345678901234567890123 zzzzzzzz mag z12345678901234567890123 zzzzzzzz (msb zero unless shifted) ph 012345678901234567 zzzzzzzzzzzzzz agc z12345678901234567 zzzzzzzzzzzzzz (msb zeroed) table 24. serial data output 1 content/format register 1 (iwa = *015h) (continued) p(31:0) function table 25. serial data output 1 content/format register 2 (iwa = *016h) p(31:0) function 31:24 set to zero. 23:16 seventh serial slot in serial data output 1 (sd1x). x = a, b, c or d. see bits 7:0 of table 24 for functional description of bits 23:16. 15:8 sixth serial slot in serial data output 1 (sd1x). x = a, b, c or d. see bits 7:0 of t able 24 for functional description of bits 15:8. 7:0 fifth serial slot in serial data output 1 (sd1x). x = a, b, c or d. see bits 7:0 of tabl e 24 for functional description of b its 7:0. HSP50216
39 table 26. serial data output 2 content/format register 1 (iwa = *017h) p(31:0) function 31:24 fourth serial slot in serial data output 2 (sd2x). x = a, b, c or d. see bits 7:0 of table 24 for functional description o f bits 23:16. 23:16 third serial slot in serial data output 2 (sd2x). x = a, b, c or d. see bits 7:0 of tabl e 24 for functional description of bits 23:16. 15:8 second serial slot in serial data output 2 (sd2x). x = a, b, c or d. see bits 7:0 of table 24 for functional description of bits 15:8. 7:0 first serial slot in serial data output 2 (sd2x). x = a, b, c or d. see bits 7:0 of tabl e 24 for functional description of b its 7:0. table 27. serial data output 2 content/format register 2 (iwa = *018h) p(31:0) function 31:24 set to zero 23:16 seventh serial slot in serial data output 2 (sd2x). x = a, b, c or d. see bits 7:0 of table 24 for functional description of bits 23:16. 15:8 sixth serial slot in serial data output 2 (sd2x). x = a, b, c or d. see bits 7:0 of t able 24 for functional description of bits 15:8. 7:0 fifth serial slot in serial data output 2 (sd2x). x = a, b, c or d. see bits 7:0 of tabl e 24 for functional description of b its 7:0. table 28. software reset register (iwa = *019h) p(15:0) function n/a writing to this location resets the followi ng activities of the f unctional block indicated. input format/select, nco, mixer and cic. clears any pending enable in each channel's input demultiplexer function, loads the ci c decimation counter (the load value is indeterminate if the decimation counter preload register has not been loaded), clears all processing enables (stops all processing in the data path, but does not clear the data path registers). filter compute engine: resets the read/write pointers, fetch instruction 31 and start the filter program execution. agc: resets the compute blocks in both the forward and loop fi lter blocks (any calculat ions in progress are lost). cartesian-to-polar coordinate converter: resets the compute blocks (any calculations in progress are lost). fifo: resets counter (clears t he fifo, all data is lost). resampler timing nco: clears the slave (active) frequency regist ers and clears the phase accumulator. output section: resets the serial output section (clears all registers, counters, and flags but does not clear the configuration registers). self test control: resets the self test control logic of the front end (input fo rmat/select, nco, mixer, and cic) and the back end (filter compute engine, agc, and cartesian-to-polar coordinate converter). table 29. channel timing advance strobe register (iwa = *01ah) p(15:0) function n/a writing to this location inserts one extra data sample in t he cic to fir path by repeating a sample. used for shifting the f ir filter compute engine timing. table 30. channel timing retard strobe register (iwa = *01bh) p(15:0) function n/a writing to this location deletes one data sample in the cic to fir path. used for shifting the fir filter compute engine tim ing. HSP50216
40 table 31. filter compute engine instruction rams (iwa = *100h thru *17fh) p(31:0) function 31:0 these locations in ram are used to store the filter compute engine instruction words. there are 128 bits per instruction wo rd with each word consisting of condition code selects, fir parameters and data r outing controls. the filter compute engine is controll ed by a simple sequencer supporting up to 32 steps where each step is defined by a 128 bit instruction word. the 128 bit instruction word is assigned to ram memory in four 32 bit data writes through t he microprocessor interface starti ng with the low 32 bits. hence, 128 32-bit memory locations are required per channel to support the 32 steps of the filter sequencer. see the filter compute engine and filter sequencer sections of the data sheet for more details. table 32. filter compute engine instruction pointer rams (iwa = *180h thru *1fch) p(15:0) function table 33. filter compute engine coefficient ram1 (iwa = *440h thru *47fh) p(31:0) function 31:0 these locations in ram are used to store the 22-bit filter coefficients used by the filter compute engine of each channel i n implementing a fir filter. the 22-bit fir filter coefficients ar e loaded in the upper 22 bits of each 32-bit ram location. the two lsbs of the second byte (bits 9:8 of the total 32 bits, 31:0) are t he shift bits. these are set to ze ro if not used. the least signi ficant byte (bits 7:0 of the total 32 bits, 31:0) are ignored. ram1 address s pace allows for storage of 64 filter coefficients out of the t otal of 192 filter coefficient storage locations. see the filter compute engi ne and filter sequencer sections of the data sheet for more de tails. table 34. filter compute engine coefficient ram2 (iwa = *480h thru *4ffh) p(31:0) function 31:0 these locations in ram are used to store the 22-bit filter coefficients used by the filter compute engine of each channel i n implementing a fir filter. the 22-bit fir filter coefficients ar e loaded in the upper 22 bits of each 32-bit ram location. the two lsbs of the second byte (bits 9:8 of the total 32 bits, 31:0) are t he shift bits. these are set to ze ro if not used. the least signi ficant byte (bits 7:0 of the total 32 bits, 31:0) are ignored. ram 2 address space allows for storage of 128 filter coefficients out of the total of 192 filter coefficient storage locations. see the filter compute engi ne and filter sequencer sections of the data sheet for more de tails. HSP50216
41 tables of global write address (gwa) registers note: these global write addresses control global functions on the HSP50216, so they are not repeated for each channel. the top five address bits select this set of registers (f8xxh). table 35. test control register (gwa = f800h) p(31:0) function 31:17 these bits can be routed to the output pins by setting bit 16 below. the bit to pin mapping is: 31 = intrpt 30 = synco 29 = serclk (unless x1 clk is selected) 28 = synca 27 = syncb 26 = syncc 25 = syncd 24 = sd1a 23 = sd1b 22 = sd1c 21 = sd1d 20 = sd2a 19 = sd2b 18 = sd2c 17 = sd2d this is provided for testing board level in terconnects. to control the serclk output, a divided down clock must be selected in the serial clock control register (gwa = f803h). 16 this bit, when high, routes bits 31:17 to the output pins in place of the normal output s. bit 0 of this register must also be set to activate this function. 15:10 unused - set to zero. 9 set-up time to clock adjust. adjusting the delay trades set up time for hold time. this bit is us ed to best center the delay w ithout a mask change. 8 set-up time to write adjust. adjusting the delay trades set up time for hold time. th is bit is used to bes t center the delay w ithout a mask change. 7:4 these bits, when set, route the msb of the sin output of the channel?s carrier nco to the number 2 serial output pin in plac e of the normal output. 7=ch0 6=ch1 5=ch2 4=ch3. 3 offset i pn by xoring bit 10 of the pn generator with the output pn. 2 enable (2 23 - 1) pn generator. the pn signal that can be added to the mixer output of each channel is produced from a (2 23 - 1) sequence, a (2 15 - 1) sequence or both. two separate generators are prov ided. the outputs of both are xored together to extend the repeat period. either or both generators can be disabled. t he xored output can further be xored with a delayed version of t he (2 23 - 1) sequence on the i channel to decorrelate it from the q channel. otherwise, the same sequence will be used on both i and q . 1 enable (2 15 - 1) pn generator. 0 test mode. when asserted, this bit puts the chip into internal (self) test mode. table 36. bus routing control register (gwa = f801h) p(31:0) function 31:24 unused - set to zero. 23:20 interrupt pulse width. the width of the interrupt pulse at the pin can be programmed to be from 1 to 15 clocks wide. progr am with the desired number of clocks. (note: the pulse counter is only rese t with the reset pin. if a channel is reset by software or a syn ci, any interrupt pulse in process will finish). 19:17 datardy delay (ch1 only). test. from 1-8. 16 ch1 or ch3 agc to ch0 ext agc. this bit selects whether the agc loop filter output from ch1 or ch3 is routed to the external agc gain input of ch0. 0=ch3, 1=ch1. 15:14 ch3 ext source mux sel. these bits select whether the ch2 s ource mux, cic2, or fir2out is routed to the external input of fir3. 0=ch2srcmux, 1=fir2, 2=cic2. 13 ch2 ext source mux sel. this bit selects whether the ch1 exter nal source mux or fir1out is routed to the external input of fi r2. 0=ch1srcmux, 1=fir1out. 12 ch1 ext source mux sel. this bit selects whether the cic0 output or fir0out is routed to the external input of fir1. 0=cic0, 1=fir0out. 11 ch0 backend input sel. 0=cic0, 1=cic1 (test). 10 ch1 backend input sel 0=cic1, 1=ch1 ext src mux. 9 ch2 backend input sel 0=cic2, 1=ch2 ext src mux. 8 ch3 backend input sel 0=cic3, 1=ch3 ext source mux. HSP50216
42 7 ch0 ext agc input enable. 0=ch0 loop filt, 1=external input. 6 ch1 ext agc input enable 0=ch1 loop filt, 1=external input. 5 ch2 ext agc input enable 0=ch2 loop filt, 1=external input. 4 ch3 ext agc input enable set to 0. 3 ch0 enable serial output 1=fir0 out enabled to serial outputs. 2 ch1 enable serial output 1=fir1 out enabled to serial outputs. 1 ch2 enable serial output 1=fir2 out enabled to serial outputs. 0 ch3 enable serial output 1=fir3 out enabled to serial outputs. table 36. bus routing control register (gwa = f801h) (continued) p(31:0) function table 37. reset/sync/interrupt source selection register (gwa = f802h) p(31:0) function 31 when set, an interrupt will be generated on each data output of c hannel 0 to the output block. ty pically, this bit will only be set for one channel. 30 when set, the data input to the part will be disabled (the input enable will be zeroed and held at zero) on a p reset (this is always true for the reset pin, whether this bit is set or not, and addi tionally, the reset pin sets the input mode to gated). the inpu t enable will be released for the input sample that aligns with the synci sig nal. this is a method for starti ng up the processing synchronous with a particular data sample. 29 when this bit is set, the carrier center frequency will be updated from the holding register (iwa = *005h) to the active regi ster on the synci signal. if the bit is set in register iwa = *004h to clear the phase accumulator feedback on loading, this function will synchronize the phase of multiple channels. after initial sy nchronization, the bit in iwa = *004h can be cleared and updates wi ll be synchronous and phase continuous across channels. 28 when this bit is set, the fir filter compute engine is reset on synci. resetting the fir filter compute engine requires 32 cl ock (clk) cycles to initialize the read and write pointers. 27 when this bit is set, the agc is reset on synci. 26 this bit has the same function as bit 29, but for the timing (resampler) nco. the bit to zero the phase accumulator feedback is in register iwa = *00ah. 25 when this bit is set, the cic decimation counter is reset on synci. 24 when this bit is set, the serial output block is reset on sy nci. if bit 4 in location gwa f803h is set, the serial clock divi der is also reset. 23:16 same functions as 31:24 for channel 1. 15:8 same functions as 31:24 for channel 2. 7:0 same functions as 31:24 for channel 3. table 38. serial clock control register (gwa = f803h) p(15:0) function 5 when set to 1, this bit will keep the se rial clock disabled after a hardware reset until receipt of the first synci signal. 4 enables resetting serial clock divider on synci. when enabled, a sy nci enabled for any of the four serial data outputs in the reset/sync register (gwa = f802h, bits 24, 16, 8 or 0) will reset the serial clock divider. 3 sclk polarity. 1 clock low to high transition occurs at the center of the data bit. 0 clock high to low transition at the center of the data bit. HSP50216
43 2:0 sclk rate. 000 serial clock disabled. 001 serial clock rate is input clk rate. 010 serial clock rate is input clk rate/2. 011 serial clock rate is input clk rate/4. 100 serial clock rate is input clk rate/8. 101 serial clock rate is input clk rate/16. other codes are undefined. table 38. serial clock control register (gwa = f803h) (continued) p(15:0) function table 39. input level detector source select/format register (gwa = f804h) p(15:0) function 15:13 channel input source selection. selects as the data input for the level detector either a(15:0), b(15:0), c(15:0), d(15:0) or the p test input register as shown below. 15:13 source selected 000 a(15:0) 001 b(15:0) 010 c(15:0) 011 d(15:0) 100 p test input register. this is provided for testing and to zero t he input data bus when a channel is not in use. the global write address register for the p test input register is f807h. 12 p register input enable select 1 = bit 11, 0 = one clock wide pulse on each write to location f808h. select 0 to write data test data into the part. select 1 to input a constant or to disable the input for minimum power di ssipation when the input level detector section is unused. 11 p input enable. when bit 12 is set, th is bit is the input enable for the p register input. active low. 0=enabled, 1=disabled. 10 parallel data input format 0 two?s complement 1 offset binary 9 fixed/floating point 0 fixed point 1 floating point. the 16-bit input bus is divided into mantissa and exponent bits group ed either 13/3 or 14/2 depending on bits 8 and 7. see text. 8:7 floating point mantissa size select. the 16-bit data input is grouped as a 13/3 or 14/2 mantissa/exponent word. these contro l bits select the mantissa/exponent grouping, add an offset to the exponent and set the shift control saturation level. 00 11/3 bits 15:5 mantissa, 2:0 exponent 01 12/3 bits 15:4 mantissa, 2:0 exponent 10 13/3 bits 15:3 mantissa, 2:0 exponent 11 14/2 bits 15:2 mantissa, 1:0 exponent 6:4 de-multiplex control. these control bits are provided to demul tiplex an input data stream comprised of a set of multiplexed data streams. up to 8 multiplexed data streams can be demultiplexed . these control bits select how many clocks after the enix signal to wait before taking the input sample. enix should be asserted for one clock period and aligned with the first channel of the multiplexed data set. for example, if four streams are multiplexed at half the clock rate, enix would align with the firs t clock period of the first stream, the second would start two clocks later, the next 4 clocks after enix , etc. the samples are aligned with enix (zero delay) at the input of the input level detector at the next enix . 000 zero delay 111 7 clock periods of delay. 3 interpolated/gated mode select 0 gated. the input level detector is updated once per clock when enix is asserted. 1 interpolated. the input level detector is upda ted every clock. the input is zeroed when enix is high. HSP50216
44 2:0 unused. set to 0. table 39. input level detector source select/format register (gwa = f804h) (continued) p(15:0) function table 40. input level detector conf iguration register (gwa = f805h) p(31:0) function 31:22 set to zero. 21 1 ones complement of 16-bit data after formatting. 0 unmodified input. 20 1 free run (ignore interval counter). 0 stop when interval counter times out. this bit may also be set low temporarily when free r unning to stabilize the accumulator data for reading. 19:18 input level detector leak factor, a. 00 1 01 2 -8 10 2 -12 11 2 -16 17:16 input level detector mode 00 leaky integrator (y n = a*x n + (1-a)*y n-1 , where a is the gain selected in bits 19:18). 01 peak detector. 10 integrator (bit 20 should be set to 0). 15:0 input level detector interval load with two less than the desired number of input samp les. the interval range is 2 to 65537 input samples. table 41. input level detector start strobe register (gwa = f806h) p(15:0) function n/a writing to this location clears the input level detector accumu lator and restarts the interval counter. when the interval co unter is done, bit 1 of the status word is set. table 42. p/test input bus register (gwa = f807h) p(15:0) function 15:0 this 16-bit value can be used as the input to one or more nco/mi xer/cic sections or to the input level detector for test or to set the input to a constant value to minimize power when the channel is not in use. the eni signal for this input is either bit 11 in the channel register at iwa *000h or the st robe generated by a write to location gwa f808h (selected via bit 12 of the channel register at iwa *000h). table 43. p/test input bus eni register (gwa = f808h) p(15:0) function n/a a write to this location, generates and eni strobe for the p driven input port (when selected via bit 12 of iwa *000h). table 44. synco strobe register (gwa = f809h) p(15:0) function n/a a write to this location will cause a one-clock-wide pulse on the synco pin. the sy nco pin is used to synchronize multiple c hannels or parts. the synco pin from one part is typi cally connected to the synci pin of all t he parts. up to two pipeline registers ma y be inserted in the synco to synci path. HSP50216
45 table of indirect read address (ira) registers the address decoding for the read source locations is given below. the internal address of the data to be read is written to di rect address 3 (add(2:0) = 3) to select and/or fetch the data. a stro be is generated, if needed, to fetch or stabilize the data for reading. if a strobe is needed, the indirect read address must be written to direct address 3 each time the data is needed. if a strobe is not needed, the data can be read repeatedly at direct addresses 0 and 1( add(2:0) = 0 and 1, respectively) with any changes in the data showing up immediately. the strobe to sample the agc gain is generated separately by an i ndirect write (see iwa *00fh in the tables of indirect write address regi sters). this allows the ag c gain of all the channels to be sampled simultaneously. note: these indirect read addresses are repeat ed for each channel. in the addresses below, the * field is the channel select nibble. these bits of the indirect address select the target channel register for the data being read. values of 0 through 3 and f are valid. table 45. p fifo read order control regist er (gwa = f820h thru f83fh) p(15:0) function 4:0 the five bits selecting the data type are encoded as follows: c c d d d, where cc is the channel number and ddd is the data type. ddd data type 000 i(23:8) the upper 16 bits of the i data path via the fifo/agc. 001 i(7:0),8*zeros the lower 8 bits of the i data path. 010 q(23:8) the upper 16 bits of the q data path via the fifo/agc. 011 q(7:0),8*zero the lower 8 bits of the q data path. 100 mag(23:8) the upper 16 bits of magnitude (a fter the gain adjust described in channel register) 101 mag(7:0),8*zero the lower 8 bits of magnitude. 110 phase(15:0) the upper 16 bits of phase. 111 agc gain (15:0) the upper 16 bits of the agc gain. table 46. table of indirect read address (ira) registers ira function *006h active carrier nco center frequency. *00ch wait preload, decr 1&2. *009h active timing nco center freq (most significant 32 bits). *00fh agc gain (must first write to agc gain read strobe register iwa = *00fh before reading). *100h - *17fh instruction rams. *180h - *1fch instruction rams (pointer dram). *400h - *43fh coefficient rom -hbf, const. *440h - *47fh coefficient ram -1. *480h - *4ffh coefficient ram -2. *500h - *5ffh coefficient rom -resampler. f806h input level detector output. HSP50216
46 absolute maximum rati ngs thermal information supply voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .+6v input, output or i/o voltage . . . . . . . . . . . . gnd -0.5v to v cc +0.5v esd classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . class iii operating conditions voltage range . . . . . . . . . . . . . . . . . . . . . . . . . . . . +3.15v to +3.45v temperature range industrial. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . -40 o c to 85 o c input low voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0v to +0.8v input high voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2v to v cc thermal resistance (typical) ja ( o c/w) 196 lead bga package (note 5). . . . . . . . . . . . . . . 27 w/200 lfm air flow . . . . . . . . . . . . . . . . . . . . . . . . . 24 w/400 lfm air flow . . . . . . . . . . . . . . . . . . . . . . . . . 23 maximum junction temperature . . . . . . . . . . . . . . . . . . . . . . .150 o c maximum storage temperature range . . . . . . . . . . -65 o c to 150 o c maximum lead temperature (soldering 10s) . . . . . . . . . . . . .300 o c caution: stresses above those listed in ?absolute maximum ratings? may cause permanent damage to the device. this is a stress o nly rating and operation of the device at these or any other conditions above those indicated in the operational sections of this specification is not implied. note: 5. ja is measured with the component mounted on a high effective ther mal conductivity test board in free air. see tech brief tb379 f or details. electrical specifications v cc = 3.3v 0.15v, t a = -40 o c to 85 o c, industrial parameter symbol test conditions min max units logical one input voltage v ih v cc = 3.45v 2.0 - v logical zero input voltage v il v cc = 3.15v - 0.8 v output high voltage v oh i oh = -2ma, v cc = 3.15v 2.6 - v output low voltage v ol i ol = 2ma, v cc = 3.15v - 0.4 v input leakage current i i v in = v cc or gnd, v cc = 3.45v -10 10 a output leakage current i o v in = v cc or gnd, v cc = 3.45v -10 10 a standby power supply current i ccsb v cc = 3.45v, outputs not loaded, no clk -500 a operating power supply current i ccop f = 70mhz, v in = v cc or gnd, v cc = 3.45v, outputs not loaded -850ma (note 6) input capacitance c in freq = 1mhz, v cc open, all measurements are referenced to device ground -7pf (note 7) output capacitance c out -7pf (note 7) notes: 6. power supply current is proportional to frequency of operation and programmed configuration of the part. typical rating for i ccop is 11ma/mhz. 7. capacitance: t a = 25 o c, controlled via design or pr ocess parameters and not directly tested. c haracterized upon initial design and at major process or design changes. electrical specifications v cc = 3.3v 0.15v, t a = -40 o c to 85 o c industrial parameter symbol min max units input and control timing clk frequency f clk -70mhz clk high t ch 5-ns clk low t cl 5-ns setup time - data inputs, input enables, synci to clk high t ds 6-ns hold time - data inputs, input enables, synci to clk high t dh 0-ns clk to output valid - synco, intrpt t pdc -6.5ns reset pulse width low t rw 5-ns reset setup time to clk high (note 8) t rs 6-ns HSP50216
47 ac test load circuit output rise, fall time (note 9) t rf -3ns microprocessor write timing p(15:0) setup time to rising edge of wr t dsw 10 - ns p(15:0) hold time from rising edge of wr t dhw -2 - ns a(1:0) setup time to rising edge of wr t asw 10 - ns a(1:0) hold time from rising edge of wr t ahw -2 - ns ce setup time to rising edge of wr t csw 10 - ns ce hold time from rising edge of wr t chw -2 - ns wr low time t wl 5-ns microprocessor read timing a(1:0) setup time to falling edge of rd t asr 8-ns a(1:0) hold time from rising edge of rd t ahr -2 - ns rd enable time t re -11.5ns rd disable time (note 9) t rd -8ns rd to p(15:0) data valid time t dv -12ns ce setup time to falling edge of rd t csr 8-ns ce hold time from rising edge of rd t chr -2 - ns serial clock output timing clk to serial data, sync and sc lk (divide-by 2 thru 16 modes) t pd -6.5ns clk low to sclk low (divide-by 1 mode, note 9) t pdl -6.5ns clk high to sclk high (divide-by 1 mode, note 9) t pdh -3ns time skew between sclk and serial data or seri al sync (divide-by 2 th ru 16 modes, note 9) t skew1 -1 1 ns time skew between sclk and serial data or serial sync (divide-by 1 mode, note 9) t skew2 0.5 2 ns notes: 8. the HSP50216 goes into reset immediately on reset going low and comes out of reset on the 4th rising edge of clk after reset goes high. 9. controlled via design or process parameters and not directly te sted. characterized upon initia l design and at major process o r design changes. electrical specifications v cc = 3.3v 0.15v, t a = -40 o c to 85 o c industrial (continued) parameter symbol min max units equivalent circuit c l (note) i oh 1.5v i ol dut switch s1 open for i ccsb and i ccop s 1 note - test head capacitance, 40pf (typ) HSP50216
48 waveforms figure 3. input and control timing figure 4. microprocessor write timing clk ain, bin, cin, din, enia , enib , enic , enid , synci synco, intrpt reset 1/f clk t ch t cl t ds t dh t pdc t rs t rw rd ce wr add(1:0) p(15:0) t dsw t dhw t asw t ahw t chw t wl t csw HSP50216
49 figure 5. micropro cessor read timing figure 6. serial output timing figure 7. output rise and fall times waveforms (continued) rd ce wr add(1:0) p(15:0) t re t dv t rd t ahr t asr t csr t chr clk sclk (/2 thru /16) sync sdxx sclk (divide by 1) t pdh t skew t pd t pdl t rf t rf 2.0v 0.5v HSP50216
50 romd fir filters - response curves figure 8. cic passband rolloff (n = # of stages, r = decimation factor, f s /r = 1 is cic output rate) figure 9. cic first alias level (n = # of stages, r = decimation factor, f s /r = 1 is cic output rate) figure 10. 5th order (n = 5) cic response (r = decimation factor, f s /r = 1 is cic output rate) note: hbf4 not included in the romd fir filter coefficient memory. see note 10 of table 48. figure 11. romd halfband filter frequency response note: hbf4 not included in the romd fir filter coefficient memory. see note 10 of table 48. figure 12. romd halfband filt er alias frequency response 0.0 -1.0 -2.0 -3.0 -4.0 -5.0 -6.0 0.00.10.20.30.40.5 f s /r db n = 5 n = 4 n = 2 n = 1 n = 3 0 -20 -40 -60 -80 -100 -120 0.00 0.10 0.20 0.30 0.40 0.50 f s /r db n = 3 n = 5 n = 2 n = 1 n = 4 -140 0 -20 -40 -60 -80 -100 -120 0.0 0.5 1.0 1.5 2.0 2.5 f s /r db -140 3.0 0 -20 -40 -60 -80 -100 -120 0 0.125 0.25 0.375 0.5 f s db hbf3 hbf4 hbf5 hbf2 -10 -30 -50 -70 -90 -110 hbf1 0 -20 -40 -60 -80 -100 -120 0 0.0625 0.125 0.1875 0.25 f s db hbf3 hbf2 hbf1 hbf4 -10 -30 -50 -70 -90 -110 hbf5 HSP50216
51 note: there is a 65db limitation in snr using the re-sampler filter. figure 13. polyphase resampler filter broadband frequency response figure 14. polyphase resampler filter pass band frequency response figure 15. polyphase resampler filter expanded resolution passband frequency response romd fir filters - response curves (continued) 0 -20 -40 -60 -80 -100 -120 magnitude (db) 12345678910111213141516 frequency (relative to f s ) 10 0 -10 -20 -30 -40 -50 -60 -70 -80 magnitude (db) 0.0625 0.125 0.1875 0.25 0.3125 0.375 0.4375 0.5 0.5625 0.625 0.6875 0.75 0.8125 0.875 0.9375 1 frequency (relative to f s ) 0 2 1 0 -1 -2 -3 -4 -5 -6 -7 -8 -9 -10 magnitude (db) 0.0625 0.125 0.1875 0.25 0.3125 0.375 0.4375 0.5 0 frequency (relative to f s ) HSP50216
52 table 47. cic passband and alias levels frequency 5th order 4th order 3 rd order 2nd order 1st order f s / r passband alias passband alias passband alias passband alias passband alias 0 0<-2000<-2000<-2000<-2000<-200 0.01 -0.007 -199.564 -0.006 -159.651 -0.004 -119.738 -0.003 -79.825 -0.001 -39.913 0.02 -0.029 -169.041 -0.023 -135.233 -0.017 -101.425 -0.011 -67.617 -0.006 -33.808 0.03 -0.064 -151.023 -0.051 -120.818 -0.039 -90.614 -0.026 -60.409 -0.013 -30.205 0.04 -0.114 -138.129 -0.091 -110.503 -0.069 -82.877 -0.046 -55.252 -0.023 -27.626 0.05 -0.179 -128.048 -0.143 -102.438 -0.107 -76.829 -0.071 -51.219 -0.036 -25.610 0.06 -0.257 -119.749 -0.206 -95.799 -0.154 -71.849 -0.103 -47.900 -0.051 -23.950 0.07 -0.351 -112.683 -0.280 -90.146 -0.210 -67.610 -0.140 -45.073 -0.070 -22.537 0.08 -0.458 -106.522 -0.367 -85.218 -0.275 -63.913 -0.183 -42.609 -0.092 -21.304 0.09 -0.580 -101.054 -0.464 -80.843 -0.348 -60.633 -0.232 -40.422 -0.116 -20.211 0.10 -0.717 -96.135 -0.573 -76.908 -0.430 -57.681 -0.287 -38.454 -0.143 -19.227 0.11 -0.868 -91.662 -0.694 -73.330 -0.521 -54.997 -0.347 -36.665 -0.174 -18.332 0.12 -1.034 -87.558 -0.827 -70.047 -0.620 -52.535 -0.413 -35.023 -0.207 -17.512 0.13 -1.214 -83.766 -0.971 -67.013 -0.728 -50.260 -0.486 -33.507 -0.243 -16.753 0.14 -1.409 -80.241 -1.127 -64.193 -0.846 -48.145 -0.564 -32.096 -0.282 -16.048 0.15 -1.619 -76.947 -1.295 -61.558 -0.972 -46.168 -0.648 -30.779 -0.324 -15.389 0.16 -1.844 -73.855 -1.475 -59.084 -1.107 -44.313 -0.738 -29.542 -0.369 -14.771 0.17 -2.084 -70.943 -1.667 -56.754 -1.251 -42.566 -0.834 -28.377 -0.417 -14.189 0.18 -2.340 -68.189 -1.872 -54.551 -1.404 -40.913 -0.936 -27.276 -0.468 -13.638 0.19 -2.610 -65.579 -2.088 -52.463 -1.566 -39.347 -1.044 -26.231 -0.522 -13.116 0.20 -2.896 -63.098 -2.317 -50.478 -1.737 -37.859 -1.158 -25.239 -0.579 -12.620 0.21 -3.197 -60.734 -2.558 -48.587 -1.918 -36.440 -1.279 -24.294 -0.639 -12.147 0.22 -3.514 -58.477 -2.811 -46.782 -2.108 -35.086 -1.406 -23.391 -0.703 -11.695 0.23 -3.847 -56.319 -3.077 -45.055 -2.308 -33.792 -1.539 -22.528 -0.769 -11.264 0.24 -4.195 -54.252 -3.356 -43.402 -2.517 -32.551 -1.678 -21.701 -0.839 -10.850 0.25 -4.560 -52.269 -3.648 -41.815 -2.736 -31.361 -1.824 -20.907 -0.912 -10.454 0.26 -4.941 -50.363 -3.953 -40.291 -2.965 -30.218 -1.976 -20.145 -0.988 -10.073 0.27 -5.338 -48.531 -4.271 -38.825 -3.203 -29.119 -2.135 -19.412 -1.068 -9.706 0.28 -5.752 -46.767 -4.602 -37.413 -3.451 -28.060 -2.301 -18.707 -1.150 -9.353 0.29 -6.183 -45.066 -4.946 -36.053 -3.710 -27.040 -2.473 -18.026 -1.237 -9.013 0.30 -6.631 -43.426 -5.305 -34.740 -3.978 -26.055 -2.652 -17.370 -1.326 -8.685 0.31 -7.096 -41.842 -5.677 -33.473 -4.257 -25.105 -2.838 -16.737 -1.419 -8.368 0.32 -7.578 -40.311 -6.063 -32.249 -4.547 -24.187 -3.031 -16.125 -1.516 -8.062 0.33 -8.078 -38.832 -6.463 -31.066 -4.847 -23.299 -3.231 -15.533 -1.616 -7.766 0.34 -8.596 -37.401 -6.877 -29.921 -5.158 -22.440 -3.439 -14.960 -1.719 -7.480 0.35 -9.133 -36.015 -7.306 -28.812 -5.480 -21.609 -3.653 -14.406 -1.827 -7.203 0.36 -9.688 -34.674 -7.750 -27.739 -5.813 -20.804 -3.875 -13.869 -1.938 -6.935 0.37 -10.262 -33.374 -8.209 -26.699 -6.157 -20.024 -4.105 -13.349 -2.052 -6.675 0.38 -10.854 -32.114 -8.684 -25.691 -6.513 -19.268 -4.342 -12.845 -2.171 -6.423 0.39 -11.467 -30.892 -9.174 -24.713 -6.880 -18.535 -4.587 -12.357 -2.293 -6.178 0.40 -12.099 -29.707 -9.679 -23.766 -7.260 -17.824 -4.840 -11.883 -2.420 -5.941 0.41 -12.752 -28.557 -10.201 -22.846 -7.651 -17.134 -5.101 -11.423 -2.550 -5.711 0.42 -13.425 -27.442 -10.740 -21.953 -8.055 -16.465 -5.370 -10.977 -2.685 -5.488 0.43 -14.119 -26.359 -11.295 -21.087 -8.472 -15.815 -5.648 -10.544 -2.824 -5.272 0.44 -14.835 -25.308 -11.868 -20.246 -8.901 -15.185 -5.934 -10.123 -2.967 -5.062 0.45 -15.573 -24.287 -12.458 -19.430 -9.344 -14.572 -6.229 -9.715 -3.115 -4.857 0.46 -16.333 -23.296 -13.066 -18.637 -9.800 -13.978 -6.533 -9.318 -3.267 -4.659 0.47 -17.116 -22.334 -13.693 -17.867 -10.270 -13.400 -6.847 -8.933 -3.423 -4.467 0.48 -17.923 -21.399 -14.339 -17.119 -10.754 -12.840 -7.169 -8.560 -3.585 -4.280 0.49 -18.754 -20.492 -15.003 -16.393 -11.253 -12.295 -7.502 -8.197 -3.751 -4.098 0.50 -19.610 -19.610 -15.688 -15.688 -11.766 -11.766 -7.844 -7.844 -3.922 -3.922 HSP50216
53 table 48. decimating halfband fir filter coefficients coeff decimating halfband #1 (dhbf #1, 7-tap) decimating halfband #2 (dhbf #2, 11-tap) decimating halfband #3 (dhbf #3, 15-tap) decimating halfband #4 (dhbf #4, 19-tap) decimating halfband #5 (dhbf #1, 23-tap) hex decimal hex decimal hex decimal hex decimal hex decimal c0 fbfe40 - 0.031303406 00c250 0.005929947 ffd538 -0.00130558 000c68 0.000378609 fff4a0 -0.000347137 c1 000000 0.000000000 000000 0.000000000 000000 0.000000000 000000 0.000000000 000000 0.000000000 c2 240100 0.281280518 f9b930 -0.049036026 0195a8 0.012379646 ff8320 -0.003810883 00525 8 0.00251 293 c3 3ffe80 0.499954224 000000 0.000000000 000000 0.000000000 000000 0.000000000 000000 0.000000000 c4 240100 0.281280518 258400 0.29309082 f83fe0 -0.06055069 0276a0 0.019245148 feb320 -0.010158539 c5 000000 0.000000000 3fff00 0.499969482 000000 0.000000000 000000 0.000000000 000000 0.000000000 c6 fbfe40 - 0.031303406 258400 0.29309082 265480 0.299453735 f70d60 -0.069904327 03e920 0.03055191 c7 000000 0.000000000 3ffe80 0.499954224 000000 0.000000000 000000 0.000000000 c8 f9b930 -0.049036026 265480 0.299453735 26ec80 0.304092407 f581a0 -0.081981659 c9 000000 0.000000000 000000 0.000000000 400000 0.500000000 000000 0.000000000 c10 00c250 0.005929947 f83fe0 -0.06055069 26ec80 0.304092407 279b00 0.309417725 c11 000000 0.000000000 000000 0.000000000 400000 0.500000000 c12 0195a8 0.012379646 f70d60 -0.069904327 279b00 0.309417725 c13 000000 0.000000000 000000 0.000000000 000000 0.000000000 c14 ffd538 -0.00130558 0276a0 0.019245148 f581a0 -0.081981659 c15 000000 0.000000000 000000 0.000000000 c16 ff8320 -0.003810883 03e920 0.03055191 c17 000000 0.000000000 000000 0.000000000 c18 000c68 0.000378609 feb320 -0.010158539 c19 000000 0.000000000 c20 00525 8 0.00251 293 c21 000000 0.000000000 c22 fff4a0 -0.000347137 notes: 10. decimating halfband filter #4 coefficients are shown for refere nce only and if it is desired to implement this fir filter th ese coefficients would have to be loaded into the fir coefficient ram (they are not included in the romd fir filter coefficient memory). 11. the 22-bit romd fir filter c oefficients are located in the upper 22 bits of the read register when read back from rom memory (except for halfband #4). these bits occupy the upper six by tes (24 bits) with the two lsbs of the lower byte (bits 9:8 of 31:0) being zero . the decimal value for the hexadecimal coefficient is calc ulated by first converting the hexadecimal value to decimal and the dividing by 2 23 (8388608). HSP50216
54 table 49. interpolating halfband fir filter coefficients coeff interpolating halfband #2 (ihbf #2, 15-tap) interpolating halfband #1 (ihbf #1, 23-tap) hex decimal hex decimal c0 ffaa24 -0.002620220 ffe944 -0.000693798 c1 000000 0.000000000 000000 0.000000000 c2 032b60 0.024761200 00a4b4 0.005026340 c3 000000 0.000000000 000000 0.000000000 c4 f07f40 -0.121116638 fd6640 -0.020317078 c5 000000 0.000000000 000000 0.000000000 c6 4cab00 0.598968506 07d240 0.061103821 c7 800000 1.000000000 000000 0.000000000 c8 4cab00 0.598968506 eb0340 -0.163963318 c9 000000 0.000000000 000000 0.000000000 c10 f07f40 -0.121116638 4f3600 0.618835449 c11 000000 0.000000000 800000 1.000000000 c12 032b60 0.024761200 4f3600 0.618835449 c13 000000 0.000000000 000000 0.000000000 c14 ffaa24 -0.002620220 eb0340 -0.163963318 c15 000000 0.000000000 c16 07d240 0.061103821 c17 000000 0.000000000 c18 fd6640 -0.020317078 c19 000000 0.000000000 c20 00a4b4 0.005026340 c21 000000 0.000000000 c22 ffe944 -0.000693798 note: 12. the 22-bit romd fir filter coefficients are located in the upper 22 bits of the read register when read back from rom memo ry. these bits occupy the upper six bytes (24 bits) with the two lsbs of the lower byte (bits 9:8 of 31:0) being zero. the decimal value for the hex adecimal coefficient is calc ulated by first converting the hexadecimal value to decim al and the dividing by 2 23 (8388608). HSP50216
55 table 50. resampler fir filter coefficients coeff hex decimal coeff hex decimal coeff hex decimal c 0 / 191 004000 0.001953125 c 32 / 159 fa3540 -0.045249939 c 64 / 127 0c2400 0.094848633 c 1 / 190 006910 0.003206253 c 33 / 158 f97f00 -0.050811768 c 65 / 126 0f8600 0.121276855 c 2 / 189 007a90 0.003740311 c 34 / 157 f8c4c0 -0.056495667 c 66 / 125 131700 0.149139404 c 3 / 188 008c90 0.004289627 c 35 / 156 f80880 -0.062240601 c 67 / 124 16d400 0.178344727 c 4 / 187 009ed0 0.004846573 c 36 / 155 f74c40 -0.067985535 c 68 / 123 1aba00 0.208801270 c 5 / 186 00b0e0 0.005397797 c 37 / 154 f691c0 -0.073677063 c 69 / 122 1ec500 0.240386963 c 6 / 185 00c230 0.005926132 c 38 / 153 f5db80 -0.079238892 c 70 / 121 22f100 0.272979736 c 7 / 184 00d240 0.006416321 c 39 / 152 f52c00 -0.084594727 c 71 / 120 273a00 0.306457520 c 8 / 183 00e090 0.006853104 c 40 / 151 f48600 -0.089660645 c 72 / 119 2b9900 0.340606689 c 9 / 182 00ecc0 0.007225037 c 41 / 150 f3ec00 -0.094360352 c 73 / 118 300a00 0.375305176 c 10 / 181 00f620 0.007511139 c 42 / 149 f36140 -0.098594666 c 74 / 117 348800 0.410400391 c 11 / 180 00fbc0 0.007682800 c 43 / 148 f2e880 -0.102279663 c 75 / 116 390c00 0.445678711 c 12 / 179 00fcb0 0.007711411 c 44 / 147 f284c0 -0.105323792 c 76 / 115 3d9100 0.480987549 c 13 / 178 00f970 0.007612228 c 45 / 146 f23980 -0.107620239 c 77 / 114 420f00 0.516082764 c 14 / 177 00eff0 0.007322311 c 46 / 145 f20940 -0.109092712 c 78 / 113 468200 0.550842285 c 15 / 176 00e050 0.006845474 c 47 / 144 f1f7c0 -0.109626770 c 79 / 112 4ae200 0.585021973 c 16 / 175 00c980 0.006149292 c 48 / 143 f20800 -0.109130859 c 80 / 111 4f2a00 0.618469238 c 17 / 174 00aad0 0.005212784 c 49 / 142 f23c80 -0.107528687 c 81 / 110 535200 0.650939941 c 18 / 173 0083b0 0.004018784 c 50 / 141 f298c0 -0.104713440 c 82 / 109 575400 0.682250977 c 19 / 172 005370 0.002546310 c 51 / 140 f31f00 -0.100616455 c 83 / 108 5b2b00 0.712249756 c 20 / 171 0019a0 0.000782013 c 52 / 139 f3d280 -0.095138550 c 84 / 107 5ed000 0.740722656 c 21 / 170 ffd590 -0.001295090 c 53 / 138 f4b500 -0.088226318 c 85 / 106 623e00 0.767517090 c 22 / 169 ff86f0 -0.003694534 c 54 / 137 f5c900 -0.079803467 c 86 / 105 656e00 0.792419434 c 23 / 168 ff2d90 -0.006422043 c 55 / 136 f71040 -0.069816589 c 87 / 104 685d00 0.815338135 c 24 / 167 fec930 -0.009485245 c 56 / 135 f88c40 -0.058219910 c 88 / 103 6b0500 0.836090088 c 25 / 166 fe59c0 -0.012886047 c 57 / 134 fa3e80 -0.044967651 c 89 / 102 6d6200 0.854553223 c 26 / 165 fddf80 -0.016616821 c 58 / 133 fc27c0 -0.030036926 c 90 / 101 6f7000 0.870605469 c 27 / 164 fd5a60 -0.020679474 c 59 / 132 fe48c0 -0.013404846 c 91 / 100 712c00 0.884155273 c 28 / 163 fccb00 -0.025054932 c 60 / 131 00a140 0.004920959 c 92 / 99 729200 0.895080566 c 29 / 162 fc31f0 -0.029726028 c 61 / 130 033140 0.024940491 c 93 / 98 73a100 0.903350830 c 30 / 161 fb9000 -0.034667969 c 62 / 129 05f7c0 0.046623230 c 94 / 97 745600 0.908874512 c 31 / 160 fae600 -0.039855957 c 63 / 128 08f400 0.069946289 c 95 / 96 74b200 0.911682129 note: 13. the 22-bit romd fir filter c oefficients are located in the upper 22 bits of the read register when read back from rom memory . these bits occupy the upper six bytes (24 bits) with the two lsbs of the lower byte (bits 9:8 of 31:0) being zero. the decimal value for t he hexadecimal coefficient is calculated by firs t converting the hexadecimal value to decimal and the dividing by 2 23 (8388608). HSP50216
56 . table 51. bit weighting for agc loop feedback path agc accum bit position gain error input gain error bit weight agc loop filter gain (mantissa) agc loop filter gain multiplier (output) agc loop filter gain (exponent) agc bit weights agc gain resolution (db) shift = 0 shift = 4 shift = 8 shift = 15 limits to output section to p 31 2222 0 0 30 2222 3 e e 48 29 2222 2 e e 24 28 2222 1 e e 12 27 15 = 2 2 2222 0 e e 6 26 14 = 1 1 2221 -1 m m 3 25 13 = 0 . 0 . 0 . 2220 . -2 m m 1.5 24 12 = 1 x 1 2 2 2 1 -3 m m 0.75 23 11 = 2 x 2 2 2 2 2 -4 m m 0.375 22 10 = 3 x 3 2 2 2 3 -5 m m 0.1875 21 9 = 4 x 4 2 2 2 4 -6 m m 0.09375 20 8 = 5 5 2 2 2 5 -7 m m 0.04688 19 7 = 6 6 2 2 1 6 -8 m m 0.02344 18 6 = 7 7 2 2 0 . 7 -9 m m 0.01172 17 5 = 8 8 2 2 1 8 -10 m m 0.00586 16 4 = 9 9 2 2 2 9 -11 m m 0.00293 15 3 = 10 10 2 1 3 10 -12 m 0.00146 14 2 = 11 11 2 0 . 4 11 m 0.000732 13 1 = 12 12 2 1 5 12 m 0.000366 12 0 = 13 13 2 2 6 13 0.000183 11 14 1 3 7 14 0.0000916 10 0 . 4 8 g 0.0000458 9 1 5 9 g 0.0000229 8 2 6 10 g 0.0000114 7 3 7 11 g 0.00000572 6 4 8 12 g 0.00000286 55913g 4 6 10 14 g 3711gg 2812gg 1913gg 01014gg HSP50216
57 all intersil products are manuf actured, assembled and tested utilizing iso90 00 quality systems. intersil corporation?s quality certifications can be viewed at website www.i ntersil.com/ design/quality intersil products are sold by description only. intersil corporation reserves the right to make changes in circuit design and/o r specifications at any time without notice. accordingly, the reader is cautioned to verify that data sheets are current before placing orders. information furnished by int ersil is believed to be accurate and reliable. how- ever, no responsibility is assumed by intersil or its subsidiaries for its use; nor for any infringements of patents or other r ights of third parties which may result from its use. no license is granted by implication or otherwise under any patent or patent rights of intersil or its subsidiaries. for information regarding intersil corporation and its products, see web site www.intersil.com sales office headquarters north america intersil corporation 2401 palm bay rd., mail stop 53-204 palm bay, fl 32905 tel: (321) 724-7000 fax: (321) 724-7240 europe intersil sa mercure center 100, rue de la fusee 1130 brussels, belgium tel: (32) 2.724.2111 fax: (32) 2.724.22.05 asia intersil ltd. 8f-2, 96, sec. 1, chien-kuo north, taipei, taiwan 104 republic of china tel: 886-2-2515-8508 fax: 886-2-2515-8369 HSP50216 plastic ball grid array packages (bga) o top view d a1 corner bottom view p n l m j k g h f e 8 13 14 12 11 10 9 corner 765 3 42 c d a b 1 side view seating plane c a a1 a2 bbb c aaa a1 corner i.d. e a b e1 d1 b s all rows and columns s m a b c c 0.15 0.08 m 0.006 0.003 a1 a1 corner i.d. e a a c v196.12x12 196 ball plastic ball grid array package symbol inches millimeters notes min max min max a - 0.059 - 1.50 - a1 0.012 0.016 0.31 0.41 - a2 0.037 0.044 0.93 1.11 - b 0.016 0.020 0.41 0.51 7 d/e 0.468 0.476 11.90 12.10 - d1/e1 0.405 0.413 10.30 10.50 - n 196 196 - e 0.032 bsc 0.80 bsc - md/me 14 x 14 14 x 14 3 bbb 0.004 0.10 - aaa 0.005 0.12 - rev. 2 12/00 notes: 1. controlling dimension: millimeter. converted inch dimensions are not necessarily exact. 2. dimensioning and tolerancing conform to asme y14.5m-1994. 3. ?md? and ?me? are the maximum ball matrix size for the ?d? and ?e? dimensions , respectively. 4. ?n? is the maximum number of balls for the specific array size. 5. primary datum c and seating plane are defined by the spher- ical crowns of the contact balls. 6. dimension ?a? includes standoff height ?a1?, package body thickness and lid or cap height ?a2?. 7. dimension ?b? is measured at the maximum ball diameter, parallel to the primary datum c. 8. pin ?a1? is marked on the top and bottom sides adjacent to a1. 9. ?s? is measured with respect to datum?s a and b and defines the position of the solder balls nearest to package center- lines. when there is an even number of balls in the outer row the value is ?s? = e/2.

▲Up To Search▲

Price & Availability of HSP50216

	To Download HSP50216 Datasheet File
If you can't view the Datasheet, Please click here to try to view without PDF Reader .