VOGONS

Operating the Tandy's SN76489 manually -- not working as expected

Topic actions

First post, by furrykef

Posted on 
User metadata
Rank Newbie
Rank
Newbie

I wasn't too sure which forum to put this in, so I'm putting it here. Please move as necessary. (Maybe it should go in one of the DosBox forums, but I'm also interested in how this behaves on a real Tandy...)

I'm writing an NSF player -- y'know, Nintendo music -- for the Tandy 1000 and IBM PCjr. So far this has worked splendidly except for one thing: the SN76489 kinda sucks at being a 2A03. Not only do the squares have no duty cycle and triangles are not possible, the bottom octave is missing. For the NES triangle channel, the bottom two octaves are missing, so many songs have no bass. I find this unacceptable.

So I thought, OK, I'll just set the frequency to zero and toggle the volume to play arbitrary waveforms manually. (This is not unlike the "RealSound" technology used for the PC speaker.) This almost works in DosBox 0.74. I wrote a test program that hooks the 1Ch (or 08h) interrupt and plays a square wave by toggling whenever the timer fires. The problem is, there are gaps in the square wave where the peak or trough is about twice as long as it should be. The result sounds like some kind of really bad arpeggio instead of a solid square wave. I have confirmed that DosBox is running literally nothing aside from my timer routine, so it's not like some other interrupt is interfering. Upon thinking about it further, I think what's happening is that the system clock is slightly out of sync with the SN76489's clock, so sometimes the volume is toggled before the SN's timer fires and sometimes it's toggled after, hence the apparent skips. However, I'm not 100% sure on this yet.

I'm trying to determine if this is a fault in DosBox or if the real hardware behaves this way. I have attached my test program below. (Note that there is no way to kill it, so it will require a reboot.) It would be nice if somebody could tell me how it sounds on a real Tandy. If a real Tandy produces a solid square, then we can work on fixing DosBox to produce the same result.

To assemble with NASM: nasm test.asm -fbin -o test.com

    cpu     8086
    org     0x100

section .text

start:
    ; Set up multiplexer
    in      al, 0x61
    or      al, 0x60
    out     0x61, al

    ; Set channel freq to 0 (flat output)
    mov     al, 0x80
    out     0xc0, al
    xor     al, al
    out     0xc0, al

    ; Silence channel
    mov     al, 0x9F
    out     0xc0, al

    ; Kill all interrupts for now
    ; (Note: in a real program, we'd need to restore them upon exit)
    mov     al, 0xff
    out     0x21, al

    ; Install timer ISR
    ; NB: In actual program, will have to uninstall ISR afterwards
    ; Remember DS == CS, so DS:DX == CS:DX
    ;mov     dx, ISR
    ;mov     ah, 0x25            ; DOS service 0x25 = install ISR
    ;mov     al, 0x08            ; 0x08 = system timer interrupt
    ;int     0x21

    ; Install timer ISR the hard way
    xor     ax, ax
    mov     es, ax
    mov     ax, ISR
    mov     [ES:8*4], ax
    mov     ax, cs
    mov     [ES:8*4+2], ax


    ; Set clock
    ; NB: In actual program, will have to reset clock when done
    ; (same code, just send 0's to 0x40 instead)
    ; Then also set DOS to real time clock
    mov     al, 0b00111010      ; @TODO@ -- which mode (bits 1-3)?
    out     0x43, al
    mov     al, 0x70            ; Low bits
    out     0x40, al
    mov     al, 4
    out     0x40, al            ; High bits

    ; Enable just the timer interrupt
    mov     al, 0xfe
    out     0x21, al

; 'Forever' is brought to you by Pinkie Pie
forever:
…Show last 32 lines
    jmp     forever


ISR:
    cli
    push    ds
    push    ax

    ; Set DS to CS
    push    cs
    pop     ds

    mov     al, [SquareToggle]
    not     al
    mov     [SquareToggle], al

    and     al, 0x0f
    or      al, 0x90
    out     0xc0, al

    pop     ds

    ; Tell 8259A interrupt done
    mov     al, 0x20
    out     0x20, al

    pop     ax
    sti
    iret


SquareToggle:   db 0

Reply 1 of 7, by ripsaw8080

Posted on 
User metadata
Rank DOSBox Author
Rank
DOSBox Author

The issue was brought up recently, so it's something of a coincidence to see a second thread on the subject:
Tandy 1000 Non-DAC Digitized PWM Samples

Reply 3 of 7, by furrykef

Posted on 
User metadata
Rank Newbie
Rank
Newbie

For all intents and purposes it's an SN76489, though. They're basically the same chip, and "SN76489" is the better known name.

I hear that the "set frequency to 0" trick doesn't work on SN chips made by TI, so it's quite possible that the test program (the ASM code above) will output nothing on a real Tandy. Only one way to find out for sure, though.

I also realized that not everybody who has a Tandy lying around is going to want to mess around with NASM. So I've attached the test program above as a prebuilt .com file. (Ironically, it's smaller unzipped than zipped, but I had to zip it because the forum doesn't like .com files.) Again, it's a Tandy .com file, so it'll only run on a Tandy.

Reply 4 of 7, by ripsaw8080

Posted on 
User metadata
Rank DOSBox Author
Rank
DOSBox Author
furrykef wrote:

I hear that the "set frequency to 0" trick doesn't work on SN chips made by TI

The Wikipedia page for the SN76489 suggests that the Tandy 1000 used an NCR clone, but perhaps not in every machine manufactured:

Another clone is the NCR 7496 or NCR 8496, used in the Tandy 1000 computer. It again has a different white noise pattern but is otherwise functionally identical to the SN76489.

Reply 5 of 7, by Cloudschatze

Posted on 
User metadata
Rank Oldbie
Rank
Oldbie
furrykef wrote:

I'm writing an NSF player -- y'know, Nintendo music -- for the Tandy 1000 and IBM PCjr. So far this has worked splendidly except for one thing: the SN76489 kinda sucks at being a 2A03. Not only do the squares have no duty cycle and triangles are not possible, the bottom octave is missing. For the NES triangle channel, the bottom two octaves are missing, so many songs have no bass. I find this unacceptable.

Tandy's "PSSJ" ASIC, found in the TL/SL and later systems, integrates the SN76496 while adding a number of enhancements. Two that may interest you include an extra bit of frequency division, allowing for frequencies down to 54.62Hz, and a DAC channel that can also function as fourth PSG voice with selectable waveform (Pulse w/variable duty cycle, Ramp, or Triangle).

The systems bearing a PSSJ chip aren't exactly uncommon, so it would be fantastic if your NSF player could utilize these extra features.

Technical and programming information can be found here:

ftp://ftp.oldskool.org/pub/tvdog/tandy1000/do … ents/tltech.zip
ftp://ftp.oldskool.org/pub/tvdog/tandy1000/so … nd/tspak181.zip

Reply 6 of 7, by Great Hierophant

Posted on 
User metadata
Rank l33t
Rank
l33t

My Tandy 1000SX uses an NCR 8496, but the tech reference schematic indicates that a TI SN76496 can be used if a resistor is changed. There is a socket for the chip.

Last edited by Great Hierophant on 2016-09-19, 16:54. Edited 1 time in total.

http://nerdlypleasures.blogspot.com/ - Nerdly Pleasures - My Retro Gaming, Computing & Tech Blog

Reply 7 of 7, by Great Hierophant

Posted on 
User metadata
Rank l33t
Rank
l33t

The test.com program produces a solid square wave tone on my SX.

http://nerdlypleasures.blogspot.com/ - Nerdly Pleasures - My Retro Gaming, Computing & Tech Blog

Go to top of page Go to top of page
Back to PC Emulation
Close window
Close window