1 tm fn2459.5 HSP48901 3 x 3 image filter the intersil HSP48901 is a high speed 9-tap fir filter which utilizes 8-bit wide data and coefficients. it can be configured as a one dimensional (1-d) 9-tap filter for a variety of signal processing applications, or as a two dimensional (2-d) filter for image processing. in the 2-d configuration, the device is ideally suited for implementing 3 x 3 kernel convolution. the 30mhz clock rate allows a large number of image sizes to be processed within the required frame time for real-time video. data is provided to the HSP48901 through the use of programmable data buffers such as the hsp9500 or any other programmable shift register. coefficient and pixel input data are 8-bit signed or unsigned integers, and the 20-bit extended output guarantees no overflow will occur during the filtering operation. there are two internal register banks for storing independent 3 x 3 filter kernels, thus, facilitating the implementation of adaptive filters and multiple filter operations on the same data. the configuration of the HSP48901 image filter is controlled through a standard microprocessor interface and all inputs and outputs are ttl compatible. features ? dc to 30mhz clock rate configurable for 1-d and 2-d correlation/convolution dual coefficient mask registers, switchable in a single clock cycle two?s complement or unsigned 8-bit input data and coefficients 20-bit extended precision output standard p interface applications image filtering edge detection/enhancement pattern matching real time video filters block diagram ordering information part number temp. range ( o c) package pkg. no. HSP48901jc-20 0 to 70 68 ld plcc n68.95 HSP48901jc-30 0 to 70 68 ld plcc n68.95 HSP48901gc-20 0 to 70 68 ld pga g68.a HSP48901gc-30 0 to 70 68 ld pga g68.a z -1 z -1 z -1 z -1 z -1 z -1 z -1 z -1 z -1 z -1 2:1 z -1 z -1 z -1 z -1 2:1 z -1 z -1 z -1 z -1 address decoder control logic clock gen clk hold 3 cin0-7 din1 (0-7) din2 (0-7) din3 (0-7) a0-2 frame ld a b c d e f g h i dout 0-19 clock internal mode mode + + + + z -1 z -1 z -1 z -1 z -1 data sheet may 1999 caution: these devices are sensitive to electrostatic discharge; follow proper ic handling procedures. 1-888-intersil or 321-724-7143 | intersil (and design) is a trademark of intersil americas inc. copyright ? intersil americas inc. 2002. all rights reserved
2 pinouts 68 lead plcc top view 68 pin grid array (pga) top view 9 8 7 6 5 4 3 2 1 6867666564636261 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 dout6 dout7 dout8 dout9 gnd dout10 dout11 dout12 dout13 dout14 dout15 dout16 dout17 dout18 dout19 v cc frames din2 (7) din2 (6) din2 (5) din2 (4) din2 (3) din2 (2) din2 (1) din2 (0) gnd din3 (7) din3 (6) din3 (5) din3 (4) din3 (3) din3 2) din3 (1) din3 (0) v cc din1 (0) din1 (1) din1 (2) din1 (3) din1 (4) din1 (5) din1 (6) din1 (7) gnd dout0 dout1 dout2 dout3 dout4 dout5 v cc v cc clk gnd cin7 cin6 cin5 cin4 cin3 cin2 cin1 cin0 gnd ld hold a2 a1 a0 11 10 9 8 7 6 5 4 3 2 1 dout6 v cc dout4 dout2 dout0 din1 (7) din1 (5) din1 (3) din1 (1) din2 (7) din2 (6) dout7 dout6 din2 (5) din2 (4) dout9 gnd din2 (3) din2 (2) dout10 dout11 din2 (1) din2 (0) dout12 dout13 gnd din2 (7) dout14 dout15 din3 (6) din3 (5) dout16 dout17 din3 (4) din3 (3) dout18 dout19 din3 (2) din3 (1) frame a2 ld cin0 cin2 cin4 cin6 gnd v cc din3 (0) a0 a1 hold gnd cin1 cin3 cin5 cin7 clk dout5 dout3 dout1 gnd din1 (4) din1 (2) din1 (0) v cc bcdefghjkl a v cc din1 (6) HSP48901
3 pin descriptions name plcc pin type description v cc 9, 27, 45, 61 the +5v power supply pins. 0.1 f capacitors between the v cc and gnd pins are recommended. gnd 18, 29, 38, 56 the device ground. clk 28 i input and system clock. operations are synchronous with the rising edge of this clock signal. din1(7-0) 1-8 i pixel data input bus #1. these inputs are used to provide 8-bit pixel data to the HSP48901. the data must be provided in a synchronous fashion, and is latched on the rising edge of the clk signal. the din1(0-7) inputs are also used to input data when operating in the 9-tap fir mode. din2(7-0) 10-17 i pixel data input bus #2. same as above. these inputs should be grounded when operating in the 1d mode. din3(7-0) 19-26 i pixel data input bus #3. same as above. these inputs should be grounded when operating in the 1d mode. cin7-0 30-37 i coefficient data input bus. this input bus is used to load the coefficient mask register(s) and the initialization register. the register to be loaded is defined by the register address bits a0-2. the cin0-7 data is loaded to the addressed register through the use of the ld input. dout19-0 46-55, 57-60, 62-67 o output data bus. this 20-bit output port is used to provide the convolution result. the result is the sum of products of the input data samples and their corresponding coefficients. frame 44 i frame is an asynchronous new frame or vertical sync input. a low on this input resets all internal circuitry except for the coefficient and int registers. thus, after a frame reset has occurred, a new frame of pixels may be c onvolved without reloading these registers. hold 40 i the hold input is used to gate the clock from all of the internal circuitry of the HSP48901. this signal is synchronous, is sampled on the rising edge of clk and takes effect on the following cycle. while this signal is active (high), the clock will have no effect on the HSP48901 and internal data will remain undisturbed. a2-0 41-43 i control register address. these lines ar e decoded to determine which register in the control logic is the destination for the data on the cin0-7 inputs. register loading is controlled by the a0-2 and ld inputs. ld 39 i load strobe. ld is used for loading the internal registers of the HSP48901. the rising edge of ld will latch the cin0-7 data into the register specified by a0-2. the address on a0-2 must be setup with respect to the falling edge of ld and must be held with respect to the rising edge of ld . HSP48901
4 functional description the HSP48901 can perform convolutio n of a 3 x 3 filter kernel with 8-bit image data. it accepts the image data in a raster scan, non-interlaced format, convolves it with the filter kernel and outputs the filtered image. the input and filter kernel data are both 8-bits, while the output data is 20 bits to prevent overflow during the convolution operation. image data is input via the din1, din2, and din3 busses. this data would normally be provided by programmable data buffer such as the hsp9501 as illustrated in the operations section of this specification. the data is then convolved with the 3 x 3 array of filter coefficients. the resu ltant output data is then stored in the output register. the HSP48901 may also be used in a one-dimensional mode. in this configuration, it functions as a 1-d 9-tap fir filter. data would be input via the din1(0-7) bus for operation in this mode. initialization of the convolver is done using the cin0-7 bus to load configuration data and the filter kernel(s). the address lines a0-2 are used to address the internal registers for initialization. the configuration data is loaded using the a0-2, cin0-7 and ld controls as address, data and write enable, respectively. this interface is compatible with standard microprocessors without the use of any additional glue logic. filtered image data is output from the convolver over the dout0-19 bus. this output bus is 20 bits wide to provide room for growth during the convolution operation. 8-bit multiplier array the multiplier array consists of nine 8 x 8 multipliers. each multiplier forms the product of a filter coefficient with a corresponding pixel in the input image. input and coefficient data may be in either two's complement or unsigned integer format. the nine coefficients form a 3 x 3 filter kernel which is multiplied by the input pixel data and summed to form a sum of products for implementation of the convolution operation as shown below: output = (a x p1) + (b x p2) + (c x p3) + (d x p4) + (e x p5) + (f x p6) + (g x p7) + (h x p8) + (i x p9) control logic the control logic (figure 1) contains the initialization register and the coefficient registers. the control logic is updated by placing data on the cin0-7 bus and using the a0-2 and ld control lines to write to the addressed register (see address decoder). all of the control logic registers are unaffected by frame . filter kernel input data a b c p1p2p3 d e f p4p5p6 g h i p7p8p9 a0 - 2 address code encro ld encr1 cas cr1 cro 3 initialization register (int) cin0 - 7 cas initialization data i0 e h0 e g0 e f0 e e0 e d0 e c0 e b0 e a0 e cr0 ihgfedcba i1 e h1 e g1 e f1 e e1 e d1 e c1 e b1 e a1 e cr1 s q c q encr1 encr0 coefficient register 1 coefficient register 0 figure 1. control logic block diagram HSP48901
5 initialization register the initialization register is used to appropriately configure the convolver for a particular application. it is loaded through the use of the cin0-7 bus along with the ld input. bit-0 defines the input data and coefficients format (unsigned or two's complement); bit-1 defines the mode of operation (1-d or 2-d); and bit-2 and bit-3 determine the type of rounding to occur on the dout0-19 bus; the complete definition of the initialization register bits is given in table 1. coefficient registers (creg0, creg1) the control logic contains two coefficient register banks, creg0 and creg1. each of these register banks is capable of storing nine 8-bit filter coefficient values (3 x 3 kernel). the output of the registers are connected to the coefficient input of the corresponding multiplier in the 3 x 3 multiplier array (designated a through i). the register bank to be used for the convolution is selectable by writing to the appropriate address (see address decoder). all registers in a given bank are enabled simultaneously, and one of the banks is always active. for most applications, only one of the register banks is necessary. the user can simply load creg0 after power up, and use it for the entire convolution operation. (creg0 is the default register). the alternate register bank allows the user to maintain two sets of filter coefficients and switch between them in real time. the coefficient masks are loaded via the cin0-7 bus by using a0-2 and ld . the selection of the particular register bank to be used in processing is also done by writing to the appropriate address (see address decoder). for example, if creg0 is being used to provide coefficients to the multipliers, creg1 can be updated at a low rate by an external processor; then, at the proper time, creg1 can be selected, so that the new coefficient mask is used to process the data. thus, no clock cycles have been lost when changing between alternate 3 x 3 filter kernels. the nine coefficients must be loaded sequentially over the cin0-7 bus from a to i. the address of creg0 or creg1 is placed on a0-2, and then the coefficients are written to the corresponding coefficient register one at a time by using the ld input. address decoder the address decoder (see figure 1) is used for writing to the control logic of the HSP48901. loading an internal register is done by selecting the destination register with the a0-2 address lines, placing the data on cin0-7, and asserting ld control line. when ld goes high, the data on cln0-7 is latched into the addressed register. the address map for the a0-2 bus is shown in table 2. while loading of the control logic registers is asynchronous to clk, the target register in the control logic is being read synchronous to the internal clock. therefore, care must be taken when modifying the convolver setup parameters during processing to avoid changing the contents of the registers near a rising edge of clk. the required setup time relative to clk is given by the specification tlcs. for example, in order to change the active coefficient register from creg0 to creg1 during an active convolution operation, a write will be performed to the address for selecting creg1 for internal processing (a0-2 = 110). in order to provide proper uninterrupted operation, ld should be deasserted at least tlcs prior to the next rising edge of clk. failure to meet this setup time may result in unpredictable results on the output of the convolver. keep in mind that this requirement applies only to the case where changes are being made in the control logic during an active convolution operation. in a typical convolver configuration routine, where the configuration data is loaded prior to the actual convolution operation, this specification would not apply. table 1. initialization register initialization register bit 0 function = input and coefficient data format 0 unsigned integer format 1 two?s complement format bit 1 function = operating mode 0 1-d 9-tap filter 1 2-d 3 x 3 filter 3 bit 2 function = output rounding 0 0 no rounding 0 1 round to 16 bits (i.e., dout19-4) 1 0 round to 8 bits (i.e., dout19-12) 1 1 not valid table 2. address maps control logic address map a2-0 function 0 0 0 reserved for future use. 0 0 1 reserved for future use. 0 1 0 load coefficient register 0 (creg0). 0 1 1 load coefficient register 1 (creg1). 1 0 0 load initialization register (int). 1 0 1 select creg0 for internal processing. 1 1 0 select creg1 for internal processing. 1 1 1 no operation. HSP48901
6 control signals hold the hold control input provides the ability to disable internal clock and stop all operations temporarily. hold is sampled on the rising edge of clk and takes effect during the following clock cycle (refer to figure 2). this signal can be used to momentarily ignore data at the input of the convolver while maintaining its current output data and operational state. frame the frame input initializes all internal flip flops and registers except for the coefficient and initialization registers. it is used as a reset between video frames and eliminates the need to reinitialize the entire HSP48901 or reload the coefficients. the registers and flip flops will remain in a reset state as long as frame is active. frame is an asynchronous input and may occur at any time. however, it must be deasserted at least t fs ns prior to the rising clock edge that is to begin operation for the next frame in order to ensure the new pixel data is properly loaded. operation a single HSP48901 can be used to perform 3 x 3 convolution on 8-bit image data. a block diagram of this configuration is shown in figure 3. the inputs of an external data buffer (such as the hsp9501) are connected to the input data in parallel with the dln1(0-7) lines; the outputs of the data buffer are connected to the din2(0-7) bus. a second external data buffer is connected between the outputs of the first buffer and the din3(0-7) inputs. to perform the convolution operation, a group of nine image pixels is multiplied by the 3 x 3 array of filter coefficients and their products are summed and sent to the output. for the example in figure 3, the pixel value in the output image at location m, n is given by: dout(m,n) = a x pm-1, n-1 + b x pm-1,n +c x pm-1, n+1 + d x pm, n-1 + e x pm, n + f x pm, n+1 + g x pm+1,n-1 + h x pm+1,n +i xpm+1, n+1 this process is continually repeated until the last pixel of the last row of the image has been input. it can then start again with the first row of the next frame. the frame pin is used to clear the internal multiplier registers and dout0-19 registers between frames. the row length of the image to be convolved is limited only by the maximum length of the external data buffers. the setup is straightforward. the user must first setup the HSP48901 by loading a new value into the initialization register. the coefficients can now be loaded one at a time from a to i via the cin0-7 coefficient bus, and the a0-2 and ld control lines. multiple filter kernels can also be used on the same image data using the dual coefficient registers creg0 and creg1. this type of filtering is used when the characteristics of the input pixel data change over the image in such a way that no one filter produces satisfactory results for the entire image. in order to filter such an image, the characteristics of the filter itself must change while the image is being processed. the HSP48901 can perform this function with the use of an external processor. the processor is used to calculate the required new filter coefficients, loads them into the coefficient register not in use, and selects the newly loaded coefficient register at the proper time. the first coefficient register can then be loaded with new coefficients in preparation for the next change. this can be carried out with no interruption in processing, provided that the new register is selected synchronous to the convolver clk signal. the HSP48901 can also operate as a one dimensional 9-tap fir filter by programming the in itialization register to 1-d mode (i.e., int bit-1 = 0). this configuration will provide for nine sequential input values to be multiplied by the coefficient values in the selected coefficient register and provide the proper filtered output. the input bus to be used when operating in this mode is the din1(0-7) inputs. the equation for the output in the 1-d 9-tap fir case becomes: d0utn = a x dn-8 + b x dn-7 + c x dn-6 + d x dn-5 + e x dn-4 + f x dn-3 + g x dn-2 + h x dn-1 + l x dn frame rate the total time to process an image is given by the formula: t = r x c/f, where: t = time to process a frame r = number of rows in the image c = number of pixels in a row f = clock rate of the HSP48901 internal clock hold clk figure 2. hold operation image data 8 20 filtered image data HSP48901 initialization data a d f b e h c f i filter kernel image data din 1 (0 -7) din 2 (0 -7) din 3 (0 -7) dout 0 -19 buffer data buffer data p m -1, n -1 p m -1, n p m -1, n +1 p m, n -1 p m, n p m, n +1 p m+1, n -1 p m+1, n p m+1, n +1 figure 3. 3 x 3 kernel on an 8-bit image HSP48901
7 absolute maximum ratings thermal information supply voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +8.0v input, output or i/o voltage applied . . . . . gnd -0.5v to v cc +0.5v esd classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . class 1 operating conditions voltage range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.75v to 5.25v temperature range . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0 o c to 70 o c thermal resistance (typical, note 1) ja ( o c/w) jc ( o c/w) plcc package. . . . . . . . . . . . . . . . . . . 43 n/a pga package. . . . . . . . . . . . . . . . . . . . 38 8 maximum junction temperature plcc package. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .150 o c pga package. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .175 o c maximum storage temperature range . . . . . . . . . . -65 o c to 150 o c maximum lead temperature (soldering 10s) . . . . . . . . . . . . .300 o c (plcc - lead tips only) die characteristics number of transistors or gates . . . . . . . . . . . . . . . . . 13,594 gates 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: 1. ja is measured with the component mounted on an evaluation pc board in free air. dc electrical specifications v cc = 5.0v 5%, t a = 0 o c to 70 o c parameter symbol test conditions min max units logical one input voltage v ih v cc = 5.25v 2.0 - v logical zero input voltage v il v cc = 4.75v - 0.8 v high level clock input v ihc v cc = 5.25v 3.0 - v low level clock input v ilc v cc = 4.75v - 0.8 v output high voltage v oh i oh = 400 a, v cc = 4.75v 2.6 - v output low voltage v ol i ol = +2.0ma, v cc = 4.75v - 0.4 v input leakage current i i v in = v cc or gnd, v cc = 5.25v -10 10 a standby power supply current i ccsb v in = v cc or gnd, v cc = 5.25v, outputs open - 500 a operating power supply current i ccop f = 20mhz, v in = v cc or gnd, v cc = 5.25v (note 2) - 120 ma input capacitance c in f = 1mhz, v cc = open, all measurements are referenced to device gnd (note 3). -10pf output capacitance c o -15pf notes: 2. power supply current is proportional to operating frequency. typical rating for i ccop is 6ma/mhz. 3. not tested, but characterized at initia l design and at major process/design changes. ac electrical specifications v cc = 5.0v 5%, t a = 0 o c to 70 o c parameter symbol notes test conditions -30 -40 units min max min max clock period t cycle 33 - 50 - ns clock pulse width high t pwh 13 - 20 - ns clock pulse width low t pwl 13 - 20 - ns data input setup time t ds 14 - 16 - ns data input hold time t dh 0-0-ns clock to data out t out - 21 - 30 ns address setup time t as 5-5-ns HSP48901
8 test load circuit address hold time t ah 2-2-ns configuration data setup time t cs 10 - 12 - ns configuration data hold time t ch 0-0-ns ld pulse width t lph 13 - 20 - ns ld setup time t lcs note 4 31 t cycle +2 40 t cycle +2 ns hold setup time t hs 10 - 12 - ns hold hold time t hh 0-0-ns frame pulse width t fpw t cycle -t cycle -ns frame setup time t fs note 5 28 - 40 - ns output rise t r from 0.8v to 2.0v - 8 - 8 ns output fall time t f from 2.0v to 0.8v - 8 - 8 ns notes: 4. this specification applies only to the case where a change in the active coefficient register is being selected during a conv olution operation. it must be met in order to achieve predictable results at the next rising clock edge. in most applications, this selection will be made asynchronously, and the t lcs specification may be disregarded. 5. while frame is asynchronous with respect to clk, it must be deasserted a minimum of t fs ns prior to the rising clock edge which is to begin loading new pixel data for the next frame. 6. ac testing is performed as follows: input levels (clk input) = 4.0v and 0v; input levels (all other inputs) = 0v to 3.0v; inp ut timing reference levels: (clk) = 2.0v, (others) = 1.5v; other timing references: v oh 1.5v, v ol 1.5v; output load test load circuit with c l = 40pf. ac electrical specifications v cc = 5.0v 5%, t a = 0 o c to 70 o c parameter symbol notes test conditions -30 -40 units min max min max dut equivalent circuit 1.5v i ol i oh (note 7) c l s 1 n otes: 7. includes stray and jig capacitance. 8. switch s 1 open for i ccsb and i ccop tests. HSP48901
9 all intersil u.s. products are manufactured, asse mbled and tested utilizing iso9000 quality systems. intersil corporation?s quality certifications can be viewed at www.intersil.com/design/quality intersil products are sold by description only. intersil corporation reserves the right to make changes in circuit design, soft ware and/or specifications at any time without notice. accordingly, the reader is cautioned to verify that data sheets are current before placing orders. information furnishe d by intersil is believed to be accurate and reliable. however, no responsibility is assumed by intersil or its subsidiaries for its use; nor for any infringements of paten ts or other rights 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 subsidiari es. for information regarding intersil corporation and its products, see www.intersil.com timing waveforms figure 4. functional timing f igure 5. configuration timing figure 6. synchronous load ti ming figure 7. hold timing figure 8. frame timing clk dout 0-19 din 0-7 t ds t dh t out t pwl t cycle t pwh a0-2 ld cin0-7 t lpw t as t ah t cs t ch clk ld t lcs hold internal clock clk t hs t hs t hh frame clk t fpw t fs HSP48901
|
