video on S3 chips/SPEA.html

Programming the SPEA board

© Attila Tarpai (tarpai76 gmail)

Previously, I was interested in hooking up ATI-TV ISA to an S3 ViRGE/DX board and display live video from a DVD player. Well, it didn't go very smooth, especially when it comes to hardware and connections. Then I've found this beautiful piece of hardware, a graphics- video-digitizer combo board built around S3 Trio64V+ and Bt819A. Here at least everything is already connected, all we have to do is programming. Well, almost.

Spea Software AG Mirage Video TV

A SPEA[1]/V7 - MIRAGE VIDEO TV, designed and built in 1994(?). Following the story line below, the DIAMOND sticker over the © Spea Software AG inscription, possibly (I don't want to remove the sticker - yet).

Analyzing the SPEA board

It's easy to see the 16-bit digital video path between S3 Trio64V+ and Bt819A. One reason I actually bought this board. (There is also TV-TUNER, Teletext decoder on it, but I save this for later programming projects. The purpose here is to get composite-video into Bt819A from a DVD-player, send it to Trio64V+ and display it.)

I took a few hi-res photos and started to discover the board's connections with a digital multimeter.

What we have:

  1. S3 Trio64V+
  2. Bt819A video digitizer
  3. GAL16V8B PLD
  4. SAA5246A Teletext decoder
  5. 32K SRAM for Teletext
  6. 2M DRAM
  8. 35.47 MHz PAL clock
  9. 28.64 MHz NTSC clock
  10. 14.8 MHz VGA clock
  11. TV-TUNER
  12. 27 MHz Teletext clock
  13. J3 34-pin LPB connector
  14. J6 header for power, INTA, RESET (Scenic/MX2?)
  15. J5 external audio header (Sonic/AD?)
  16. 74HCT125D Quad buffer/line driver
  17. CD4052B channel multiplexer (for I2C)
  18. CD4052B channel multiplexer (for audio)
  19. NE555D Precision Timer
  20. TDA1308 stereo headphone driver
  21. VGA out
  22. audio out
  23. Composite/S-video in
  24. Cable in

The circled area is the 16-bit data path between Bt819A and Trio64V+.


GAL16V8B is a programmable logic device from Lattice Semiconductor. That was bad news for my project: exactly what was obscure got even more obscure. After identifying pins and connections it seemed like the purpose of the logic is to drive VS/HS/LCLK for Trio64V+ based on a bunch of video timing inputs from Bt819.

GAL in/out

  --> HRESET   
  --> VRESET                                Example wiring for HS
  --> ACTIVE
  --> FIELD                                      . . . . . . . . . . . . . . . .  LPB 
  --> DVALID                                     . . . . . . . . . . . . . . . . connector
  --> QCLK                                           |                           
  --> CLKx1                                          +-- 4.7KΩ -- Vcc             
  <-- OE                                          HS |                           
Trio                                                 +-- 470Ω -- GAL -- Bt        
  --> HS                                             |                           
  --> VS                                             |                           
  --> LCLK                                           +-- 470Ω -- Trio             
  <-- GOP0
  <-- GOP1
 --> INH

The GAL also drives Bt819A's OE# pin, which is not active after reset. The 2 inputs from Trio64V+, GOP0 and GOP1 are high after reset. I tried all combinations and GAL kicks in when GOP_1_0 = 10: it asserts OE#, the Bt819A timing- and data pins emerge from 3-state, then GAL starts to drive HS/VS/LCLK for Trio64V+. This is how to switch on Bt819A.

In the meantime I bought a Saleae logic analyzer to investigate GAL and the HS/VS/CLCK output.


When it comes to fields of video data the results were interesting: GAL outputs VS @ 30Hz (and not 60Hz), skipping every other field, while Bt819A outputs individual fields separated by VRESET:

Spea solved the de-interlacing problem by simply dropping every even field sent to Trio64V+, which means also losing half of vertical resolution. An elegant solution: 30 x 240-lines half-frames per sec gives nice motion, enough resolution (we're in 1995) and are better than those horrible interlace-combs. Trio64V+ has no de-interlacing function at all (apart from that LPB can also skip every other fields, see MMFF00_25?).

When the LPB window is increased vertically beyond 240 lines, the even field reveals at the bottom and is different from the odd field. Video out froma a Canon Powershot A590 connected to Spea composite-in:

Bypassing the GAL and interlaced

I still wanted to get full vertical resolution out of Bt819A somehow, but from the GAL's VS-signal it's impossible to interlace the two fields for Trio64V+. So I experimented bypassing the GAL. There was one big problem: only GAL can assert OE# for Bt819A, but GAL must be enabled for this, setting GOP[1..0]= 10 and will also drive HS/VS/CLCK lines in the same time. I really didn't want to, but desoldered the OE# pin..

Then with GAL disabled I tried many Bt819A video timing pins to connect to Trio64V+: HRESET, ACTIVE, VRESET, DVALID, CLKx1 and QCLK right from Bt819A. I used GAL-pins as connection-points, holes on the board, and many trials changing the code.

The Trio64V+ code has been changed to use double-buffers and fill every other lines after VS occurence (VRESET → VS). Horizontal scaling is the same, around 330 pixels. The LPB window height is the same (240 lines), but the FB filled interlaced from alternating fields:

Not bad. For still pictures. The problem with alternating fields is that there is no way to know which one is odd or even for Trio64V+. When lucky, we interlace the right ones. When unlucky, we interlace opposite. The last line of the odd field ends in the middle, and the even field starts with a half-line. This will most likely confuse Trio64V+ and messes up the screen (skips the first line or the first line runs beyond the right edge).

I guess that's why results varied, clearly seen on this paused DVD frame. Depending on when the first VS came, Trio64V+ locked to even or odd field as the first:
The good (lucky one): The unlucky one:

      odd           even          odd                         odd           even          odd           even    
   __________    __________    __________                  __________    __________    __________    __________ 
  | -------- |  |     ~~~~ |  | -------- |                | -------- |  |     ~~~~ |  | -------- |  |     ~~~~ |
  | -------- |  | ~~~~~~~~ |  | -------- |                | -------- |  | ~~~~~~~~ |  | -------- |  | ~~~~~~~~ |
  |_---______|  |_~~~~~~~~_|  |_---______|                |_---______|  |_~~~~~~~~_|  |_---______|  |_~~~~~~~~_|
       |             |                                                       |             |     
       +-------------+                                                       +-------------+     
         _____|____                                                            _____|____        
        |     ~~~~ |                                                          | -------- |       
        | -------- |   proper                                                 |     ~~~~ |  will be     
        | ~~~~~~~~ |   interlacing                                            | -------- |  wrong
        | -------- |                                                          | ~~~~~~~~ |       
        | ~~~~~~~~ |                                                          | ---      |       
        | ---      |                                                          | ~~~~~~~~ |       
        |__________|                                                          |__________|       

The analog video signal does differentiates between the two: the signal pattern in the vertical synchronization period is different. Bt819A indicates field parity on the FIELD pin, but Trio64V+ cannot make any use of it - unfortunately.

When playing the DVD, well, I can confirm that the interlace-artifacts were awful..

NE555D Precision Timer

"These devices are precision timing circuits capable of producing accurate time delays or oscillation." (Ti, STMicroelectronics, STMicroelectronics)


  1. ^  Spea Software AG A company from Bayern, Germany, established in 1985. Buys Video Seven, US CA, in May 1993 (SPEA/V7 products). Videoseven has been developing and marketing graphics cards and software since 1983. Video Seven is bankrupt in the same year, July 1993. Diamond Multimedia obtains Spea in November 1995.

Mon Feb 11 07:57:02 UTC+0100 2019 © A. Tarpai