Dirt cheap frame grabber

This is a story of a a video digitizing circuit I built early 1990′s. It kind of worked, but was not practical then and much less useful nowadays. It is an interesting how simple circuit could be used for converting analogue composite video signal to digital for computer processing.

This is the story of Dirt cheap frame grabber (DCFG). The purpose of the DCFG is to simply provide a very simple method to grab video pictures and display them on a computer monitor. This device design is free and was released to the public domain. The DCFG required at least a 80286 or better and a VGA display to run.

Here is the built circuit in a small plastic box and wire that goes to PC parallel port interface.

Here is what is inside the box:

And here is the circuit diagram:

                    Dirt Cheap Frame Grabber
    (DB25P)                                     
    DO0 ---330ohm-------------*--->|--*-->|---*    IC1 = LM339   
    pin 2                  Vr |  red    1N914 |          quad   
                       (2.4v) |  led          |        comparator
    DO1 ---330ohm-------------*               |                   
    pin 3                     |               |    * = connection    
                             330ohm           |                       
          1N914               |               |    ) = no connect
    DO6 ---|<---*---1Kohm-----* Vc (2.21v)    |       (jump over)
    pin 8                     |               | 
          1N914               |               | Vr = ref voltage  
    DO7 ---|<---*---510ohm----*------*        |      2.4 volts        
    pin 9                     |      |        |                   
                   (trimmer) 5Kohm<--*        | Vc = 00 = 1.65v   
                              |               |      01 = 1.80v   
    DO2 ----------*-----------)------| |------*      10 = 1.95v   
    pin 4         |           |     .1uf      |      11 = 2.21v   
                 3|           |               |                       
    DC3           |/ |6       |               | Vz = 00 = 0.74v   
    (slct)     1 /  -|--------* Vz (1.0v)     |      01 = 0.81v   
    DS6 --------|IC1a|        |               |      10 = 0.88v   
    (ack)        \  +|----*   |               |      11 = 1.00v   
    pins 10,17     \ |7   |  360ohm           |                     
                          |   |               | Vy = 00 = 0.45v   
    DC2            / |4   |   |               |      01 = 0.54v     
    (init)     2 /  -|----)---* Vy (0.66v)    |      10 = 0.59v   
    DS5 --------|IC1b|    |   |               |      11 = 0.66v    
    (pe)         \  +|----*   |               |                   
    pins 12,16     \ |5   |  220ohm           | Vx = 00 = 0.34v   
                          |   |               |      01 = 0.37v        
    DC1            / |10  |   |               |      10 = 0.40v   
    (afd)     13 /  -|----)---* Vx (0.45v)    |      11 = 0.45v   
    DS4 --------|IC1c|    |   |               |                   
    (slct)       \  +|----*   |               | Vs = 00 = 0.086v  
    pins 13,14     \ |11  |  360ohm           |      01 = 0.093v  
                          |   |               |      10 = 0.101v  
    DC0            / |8   |   |               |      11 = 0.115v  
    (stb)     14 /  -|----)---* Vs (0.12v)    |  Assumes Vz set
    DS7 --------|IC1d|    |   |               |  for 1.0v using 
    (busy)       \  +|----*   |               |  5Kohm trimmer 
    pins 11,1   12|\ |9   |  120ohm           |  with DO6, DO7 
                  |       |   |               |  set high.      
    Gnd ----------*-------)---*---------------*------*--------*
    pins                  |   |               |      |        |
    18,19,20,21,          |  100Kohm          |     75ohm     |
    22,23,24,25           |   |      .1uf     |      |        |
                          *---*-------| |-----)------*------o )          
                   2N2222 |                   |             RCA   
            *---------C   E                   |            phono
            |           B                     |            jack 
    pin 5   |           |      1N914          | 
    DO3 ----*--4.7Kohm--*-------->|-----------*

This circuit consists of four fast comparators. They compare the video signal voltage levels to the the given reference voltage levels. The reference voltage levels are set so, that the circuit can detect four video signal gray scale levels and video sync signal. Those signals are fed to the PC parallel port digital input pins. The circuit takes power from the parallel power data lines (that are set to logic high to provide needed power).

The Dirt Cheap Frame Grabber needs to be connected the computer with a short cable. We are dealing with high speed signals here (up to 4MHz). It requires as short a connection to the computer as you can give it. Here I had around 20 cm long wires.

The DCFG hardware only provides 4 levels of gray scale. From DCFG version 2.01 the software can provide up 12 gray scales by reading the port four times for each display pixel and interpolating the results.

Because video signal is very fast changing signal, it needs to be digitized at fast rate. The heart of the program relies on a very fast IO port to memory instruction. It can operate as fast as the computer can shove data from the port to the memory. This is as fast as DMAing the data could do. The downside is that this method is sensitive to computer speed, and you might need to adjust the horizontal sample size to get viewable picture.

Because of the simple technology and not very accurate sampling, you will probably notice some sparkles (pixels shifting left and right one pixel position). This is caused by sample jitter. The jitter is caused by the fact that the sampler is not perfectly synchronized to the incoming video signal.

To use the Dirt Cheap Frame Grabber, plug it into a PC printer port, and attach the other end to the video output from a VCR or TV Monitor and run the program. Use the left or right arrow keys to adjust the horizontal sample size to obtain a stable picture. Years ago I got this working. I got some small low quality video signal on the computer screen.

This circuit is pretty useless nowadays, when you can buy good quality commercial video digitizing hardware that plugs to USB port for very little money (less than $10 for composite video and HDMI capture for some more money). Other things are that the legacy parallel port are pretty died hardware family, and things like USB-parallel port adapters and different virtualization software approaches will not cut in for this applications.

If you want to explore nowadays to get this hardware to work with some modern PCs for any reason, the best cheap idea could be to connect this hardware to a logic analyzer that can stream the data quickly in real time to PC. For example sigrok software supports lots of hardware, including cheap Chinese clones of some USB logic analyzer hardware. This kind of logic analyzers can record up to 8 channels at up to 24MHz sample rate through USB port to PC. sigrok software can show those signals, and supports you to write your own signal decoder modules that can convert the recorded bit stream to whatever format needed (or example to images). Theoretically I could plug in this DCFG circuit to logic analyzer and then decode video from recorded data. Maybe not worth of the trouble, but should be doable I think.

Here is the original Dirt Cheap Frame Grabber documentation

Dirt Cheap Frame Grabber V2.02
by Michael Day
as of 5 January 1992
Public Domain

This device is free and released to the public domain.
If you actually find a use for it, all the more power to ya!

The purpose of the DCFG is to simply provide a very simple method
to grab video pictures and display them on a computer monitor.

Version 1.01 of the DCFG only provided 4 levels of gray scale.
Version 2.01 makes a modification which allows for 12 gray scales
by reading the port four times and interpolating the results.
Version 2.02 fixes a schematic error (DO3 is pin 5, not pin 4).

The DCFG requires at least a 80286 or better and a VGA
display to run. While it is possible to change the program to
allow operation on a 8086 and/or a different display, I’m not
doing it. If you want to, you can. My intent was to just prove
that it was possible.

The frame grabber program at this point is very simple, and as a
result has some problems. The heart of the program relies on a
very fast IO port to memory instruction. It can operate as fast
as the computer can shove data from the port to the memory. This
is as fast as DMAing the data could do. This method is sensitive
to computer speed. As such, you may not get a stable picture at
first. If you don’t, use the left or right arrow keys to adjust
the horizontal sample size. The up/down arrows adjust the
vertical sample size, but this normally doesn’t need to be done.

You will probably notice some sparkles (pixels shifting left and
right one pixel position). This is caused by sample jitter and is
normal. All sampling systems have this problem. Some less than
others. Even your TV has the problem, but it is normally good
enough that you don’t see it.

The jitter is caused by the fact that the sampler is not
perfectly synchronized to the incoming video signal. Your TV has
a lot of circuitry in it to achieve the synchronization. Being
that this is a dirt cheap frame grabber, there is only limited
synchronization of the picture, and no synchronization of the
sampler. If you want to try to improve on it, please do.

To use the Dirt Cheap Frame Grabber, plug it into a PC printer
port, and attach the other end to the video output from a VCR or
TV Monitor and run the program. Use the left or right arrow keys
to adjust the horizontal sample size to obtain a stable picture.

Do not attach the Dirt Cheap Frame Grabber to the computer with a
long printer cable. We are dealing with high speed signals here
(up to 4MHz). It requires as short a connection to the computer
as you can give it. I built my circuit inside the back shell of
the connector that plugs into the printer port on the computer.

                    Dirt Cheap Frame Grabber
    (DB25P)                                     
    DO0 ---330ohm-------------*--->|--*-->|---*    IC1 = LM339   
    pin 2                  Vr |  red    1N914 |          quad    
                       (2.4v) |  led          |        comparator
    DO1 ---330ohm-------------*               |                   
    pin 3                     |               |    * = connection    
                             330ohm           |                       
          1N914               |               |    ) = no connect
    DO6 ---|<---*---1Kohm-----* Vc (2.21v)    |       (jump over)
    pin 8                     |               | 
          1N914               |               | Vr = ref voltage  
    DO7 ---|<---*---510ohm----*------*        |      2.4 volts        
    pin 9                     |      |        |                   
                   (trimmer) 5Kohm<--*        | Vc = 00 = 1.65v   
                              |               |      01 = 1.80v   
    DO2 ----------*-----------)------| |------*      10 = 1.95v   
    pin 4         |           |     .1uf      |      11 = 2.21v   
                 3|           |               |                       
    DC3           |/ |6       |               | Vz = 00 = 0.74v   
    (slct)     1 /  -|--------* Vz (1.0v)     |      01 = 0.81v   
    DS6 --------|IC1a|        |               |      10 = 0.88v   
    (ack)        \  +|----*   |               |      11 = 1.00v   
    pins 10,17     \ |7   |  360ohm           |                     
                          |   |               | Vy = 00 = 0.45v   
    DC2            / |4   |   |               |      01 = 0.54v     
    (init)     2 /  -|----)---* Vy (0.66v)    |      10 = 0.59v   
    DS5 --------|IC1b|    |   |               |      11 = 0.66v    
    (pe)         \  +|----*   |               |                   
    pins 12,16     \ |5   |  220ohm           | Vx = 00 = 0.34v   
                          |   |               |      01 = 0.37v        
    DC1            / |10  |   |               |      10 = 0.40v   
    (afd)     13 /  -|----)---* Vx (0.45v)    |      11 = 0.45v   
    DS4 --------|IC1c|    |   |               |                   
    (slct)       \  +|----*   |               | Vs = 00 = 0.086v  
    pins 13,14     \ |11  |  360ohm           |      01 = 0.093v  
                          |   |               |      10 = 0.101v  
    DC0            / |8   |   |               |      11 = 0.115v  
    (stb)     14 /  -|----)---* Vs (0.12v)    |  Assumes Vz set
    DS7 --------|IC1d|    |   |               |  for 1.0v using 
    (busy)       \  +|----*   |               |  5Kohm trimmer 
    pins 11,1   12|\ |9   |  120ohm           |  with DO6, DO7 
                  |       |   |               |  set high.      
    Gnd ----------*-------)---*---------------*------*--------*
    pins                  |   |               |      |        |
    18,19,20,21,          |  100Kohm          |     75ohm     |
    22,23,24,25           |   |      .1uf     |      |        |
                          *---*-------| |-----)------*------o )          
                   2N2222 |                   |             RCA   
            *---------C   E                   |            phono
            |           B                     |            jack 
    pin 5   |           |      1N914          | 
    DO3 ----*--4.7Kohm--*-------->|-----------*

Examplesº DCFG² 1 <-­ loadó prograí foò uså witè LPT1. DCFG² 2 <-­ loadó prograí foò uså witè LPT2. Iæ yoõ don'ô givå iô á number¬ thå prograí wilì defaulô tï LPT1® Grabbing a Frame: The program works by grabbing samples from the printer port in frames. Once each time through the loop, it sucks in a video frame from the video input. Actually, more than a frame is grabbed. How much is grabbed depends on the speed of access to the printer port on your computer. On a typical computer this is 1.5us per byte. At the 1.5us access rate, 63.5us per scan line, and 265.5 scan lines per video frame, that results in 11240 bytes per frame required to be collected. The collection size is currently set to 20,000 bytes. That provides enough space to handle buss speeds up to 12MHz which should be adequate for most computers. The down side is that we collect 210 extra scan lines on a normal computer. In most situations this isn't a problem since three to four frames will typically be lost while the collected video data is being processed anyway. The extra partial frame has only a slight effect on the over all collection time and results. The effect is minimized on a fast machine since nearly the entire array is filled anyway. The maximum sample size allowed is 65520. More than that and we run out of segment space in memory. Also during this period I keep the system interrupts off so as to not have distortions of the collected data caused by the system deciding to go off and do something else. In addition to the collection time, there is a pre-configure time while the loop is waiting for the vertical sync pulse to come in. I don't collect any data prior to the vertical sync pulse since it would be ignored anyway. The vertical sync is determined by watching the sync line. When it goes on for longer than 10us, you are in the vertical sync time. Less then 10us, and it is a horizontal sync pulse. Once the video data is collected, I release the interrupts to allow the computer to return to normal operation. Should no sync come in during the vertical sync detection period, I terminate the loop after a while to allow the computer to continue to operate, and return an error condition so that the program will know that no data was collected. If you have an O'scope, you can verify the video collection time by observing data output line DO7 (pin 9). This line should never go high for longer than 50ms. Typically it should be around 20 to 40ms in length. This should be the same whether you have a signal going into the video line or not. Though there may be a difference between the timing when a signal is present verses when it is not present. Once I get the data, I convert it to in intermediate form to be stored in the interpretation array. This array keeps the gray level values collected on the previous frames. The display routine will later interpret the values stored in the interpretation array into an absolute gray scale level for each display pixel. The video data is converted by searching for the horizontal sync pulse. Once found, we search for the end of the sync pulse. Following the end of the sync pulse is the actual video data. Each collected byte in the scan line is read and converted to the intermediate gray scale level. There are only four levels involved; 00, 01, 10, and 11. Note that since sync is not video data, it is not included in the gray scale level. Even if it were included, it would be a 00 value. Once converted, the byte is then shifted into it's proper frame position in the interp array byte and stored in the array. This is repeated until either the full width of the interpretation array is filled, or a new sync pulse is encountered. Should a sync pulse be encountered, the rest of the interpretation array values are filled with 00. Should the full width of the interpretation array be encountered, then the rest of the scan line is discarded until an new sync pulse is encountered. This sequence is continued until the entire frame has been converted. Displaying the Data: Once the frame has been converted, the interpretation array is then displayed. Each byte is read from the interpretation array and translated to an absolute gray level using the translation array located in the first 256 bytes of the interpretation array. There are 12 levels of gray available, plus black. The translation array was copied into the interpretation array at the time the interpretation array was created. conversion chart: gray level interpretation chart frame data gray F3 F2 F1 F0 F3 = frame 3, F2 = frame 2 level: 76 54 32 10 F1 = frame 1, F0 = frame 0 12: 11 xx xx xx each group of two bits 11: <11 11 xx xx represent the video level 10: <11 <11 11 xx for the frame indicated 9: <11 <11 <11 11 8: 10 <11 <11 <11 xx = any bit pattern 7: <10 10 <11 <11 <11 = less than 11; (10, 01, 00) 6: <10 <10 10 <11 <10 = less than 10; (01 or 00) 5: <10 <10 <10 10 11, 10, 01, or 00 = the indicated 4: 01 <10 <10 <10 absolute bit pattern 3: 00 01 <10 <10 2: 00 00 01 <10 the gray level for the specified 1: 00 00 00 01 bit pattern is shown at the left 0: 00 00 00 00 There are no sync bytes in the interpretation array since they were stripped out during the process of converting the video data to the interpretation data. Note that it takes four video frames to completely update the interpretation array. However, the actual displayed data will reflect the highest gray level determined from the four previous reads. This means that if the video level shifts towards the white level, the change on the display will be immediate. If the video level moves towards the black level, the change on the display will take up to four frame grabs to reflect the lower video level for that pixel. One big problem is that since the number of scan lines collected is much greater than the resolution of pixels collected in each scan line, the aspect ratio of the picture would be out of whack if we were to display the data pixel for pixel as we received it. In order to get the aspect ratio back to a reasonable value, we must throw data away. We do this by discarding three scan lines for every one that we show. If you have a very fast buss, you may want to change this to discard two scan lines instead of three in order to get a more reasonable aspect ratio. The other thing I do is to pull a little trick to improve the horizontal resolution. Normally there is a typical horizontal resolution of about 40 pixels. This is not very much, and it would be useful if there were a better resolution. There is a halfway solution. What I do is display the even pixels on a scan line from the scan line that I'm currently processing. I then display the odd pixels from the next scan line on the succeeding video frame. What this does is to provide a random selection on the odd pixels between the normal two sequential scanned pixels that effectively improves the display resolution by about 50%. hat it worsens the jitter/twinkle problem since the odd pixel is developed from a pseudo-random sampling mechanism which is not consistent. One thing to keep in mind is that the frame grab routine normally takes between one and three video frames to collect data, and during this time the interrupts are disabled. What that means is that if you are operating on a computer which has a clock interrupt faster than the normal 55ms, you will lose time, and may experience problems. Also note that since the interrupts are disabled, anything that uses high speed interrupts during the time the video data is being collected (such as RS232), will not work. The interrupts will be lost. The main factor that affects speed on slower machines are the conversion and display routines. The INS instruction used to collect the video data is relatively cpu performance independent. The main thing that affects its operation is the buss performance of the computer. The faster the buss, the more video pixels that can be collected per scan line. One possible consideration on slower machines would be to discard the extra scan lines in the Video to interpretation array conversion rather than in the interpretation to display routine. That way the extra scan lines would not even be processed, thus increasing the speed of the overall process loop. --- Future considerations for someone to do if they wish to improve on the design would be to improve the synchronization to reduce the jitter problem. At this point I don't have any ideas on how to improve the problem. Another improvement that is possible to do would be to analyze the captured video frame to determine the proper values to use for capturing the data such as scan line length in pixels, vertical sync length, total scan lines, etc. A mechanism to increase the size of the displayed picture, and maintain the best resolution possible while doing so. Save an image to disk or a series of images to make a movie. Be able to print a captured image. Integrate the image with sound. And of course integrating this stuff into Windows would be nice. That's all folks!
###########################################################################