Doug,
You could also do some kind synthesizer using fixed frequencies and
pulse width modulation.
It would not be that difficult to add attack and decay and even be
polyphonic.
You could either use the keyboard for single tone with different keys to
change the PWM, Attack, Decay, white noise, pink noise, and volume of
each channel or you could write some kind of interpreter that would play
music from some kind of text file.
Way back in the 80's I was able to do stereo 4 part harmony on a 2 MHZ
6809 using two 8-bit D/A converters.
If you want to generate different waveforms they can be stored in memory
and the faster you cycle through them the higher the frequency.
However, the finer the resolution the greater the memory requirement.
Square waves are the easiest to produce as they don't need waveform
tables. Just remember you need to remember the Nyquist frequency. You
need to be writing to the D/A at at least double the frequency of the
tone you are trying to product. However, twice the frequency only
allows you to produce square waves.
I would look at the output of your program on an oscilloscope to see if
there is any high frequency artifacts that you might want to filter out
electronically.
Have fun and please post some recordings.
On 7/9/2023 10:07 PM, Martin Bishop via cctalk wrote:
Doug
You have the essence of what it can do and how it might do it, an old iron
PDP-11 / LSI-11 can't organically do the signal processing necessary to go from
a compact signal representation to audio. That said you could use a one bit
output, which would for efficiency require a serialiser, to drive a sigma delta
converter, e.g.
https://digilent.com/shop/pmod-i2s2-stereo-audio-input-and-output/, to convert
a word stream to sound. Generating the I2S bit stream is of course an XFU,
probably doable from a DRV11 type interface, with a fair bit of logic or your
favorite microcontroller as intermediary.
A 7 kHz sample rate will not be easily achieved, unless you can use an external clock or plug in an
"optimal" oscillator, e.g. the classic example (but not for this requirement) is one
running at a multiple of baud rate frequencies, 12M288 Hz say. On these old boards, frequency =
oscillator / prescaler / divisor. Generally, the divisor is the "round" number and the
frequency is the fractional mess. Also, the prescalers may be powers of two, ten, or 1/2/5 : YMMV.
The sort of thing you could demonstrate as audio output from your hardware is
either Morse or audio composed of phonemes. The Morse would require some
evolution of the test generator I described and a little tapering of the attack
/ decay of the di dah dit sounds to avoid harshness, or could be done from
canned waveforms (returning to 0V). The phonemes might generate text to speech
of 1980's (i.e. contemporary period) standard.
Have fun
Martin
-----Original Message-----
From: Douglas Taylor via cctalk [mailto:[email protected]]
Sent: 10 July 2023 02:20
To: Martin Bishop via cctalk <[email protected]>
Cc: Douglas Taylor <[email protected]>
Subject: [cctalk] Re: Talking PDP11
Wow! Actual engineers responding...
It looks like I could only do the most rudimentary audio.
1. Sample Rate: You got maybe 20K samples to store in lower memory. At 7KHz
sample rate that would allow 3 seconds of audio. Voice only.
2. Samples: They must be 12 bits. Converting a modern audio clip requires, band
filtering, resampling and mapping to 12 bit integers. Could be done in python,
they have libraries.
3. Clocking output: I have a KMV11, but never programmed around it.
4. Amplify output: AAV11-C produces -10 to +10 volts, have to divide this down
for input to an audio amp.
In the end I will have undone all the advances made in digital audio in the
last 30 to 40 years.
Doug
On 7/9/2023 4:09 PM, Martin Bishop via cctalk wrote:
You just did use it to play "audio" :<)
The 6 us settling time corresponds to a sampling rate of ~167 kHz, not that you
will ever get there or would wish to.
The theoretical (real) sampling rate required for a given bandwith is
Fs = 2 Bw. That requires brick wall filters and it is a lot of work
to get close without significant distortion. These old DACs are all
but certain to use ladder circuits [see e.g.
https://en.wikipedia.org/wiki/Resistor_ladder] the settlng time will
mostly come from the output buffer [see e.g.
https://www.analog.com/en/technical-articles/ltc1668-dac-lt1807-opamp-
achieve-90ns-settling-to-16bits-83db-sfdr-small-footprint.html for
bleeding edge examples]. To see something other than ringing on a
scope you want at least 10 samples per cycle, e.g. for 3 kHz bandwidth
(i.e. 0 - 3 kHz frequency coverage) output at 30 kHz or greater. A
low pass (reconstruction in the argot) filter will round off the
corners - set the corner just above the passband
DMA, a local FIFO or at least double buffering are the minimum to
avoid sample jitter. On basic hardware you will probably have to do
what you can with a sampling clock derived from the RTC card, from 10
MHz you could get an interupt at 40 kHz or 25 kHz but maybe not 30
kHz. The interupt then controls the play out from a table or disk ;<)
For testing you can do quite a lot with a single cycle sine wave table in
memory. Say you are playing out at Fs = 30 kHz, and you have a 30 k sample
table. By varying the step through the table from 15k to 1 you can alter the
output frequency from 15 kHz to 1 Hz in 1 Hz increments; i.e. output frequency
= Fs * stride / table length.
From a VQ look at the AAV11 docs it uses the bottom 12 bits, doubtless <11>
is ms, hopefully it will like 2's complement numbers and the analog offset voltage
will be trimmed for bipolar signals.
Have fun and good luck
Martin
-----Original Message-----
From: Douglas Taylor via cctalk [mailto:[email protected]]
Sent: 09 July 2023 19:46
To: General Discussion: On-Topic and Off-Topic Posts
<[email protected]>
Cc: Douglas Taylor <[email protected]>
Subject: [cctalk] Talking PDP11
I have a PDP-11/53 and have just started playing with an AAV11-C D/A board. It
is a 4 channel D/A convertor with 12 bit resolution.
Can it be used to play an audio bit stream?
Here is simple code used to see if the thing was actually working:
.title AAV11 D/A test
;
.asect
dbr0 = 170440
.=1000
start:
mov #7777,r0 4096 value to R0
mov #dbr0,r1 first D/A buffer out
loop: mov r0,(r1) transfer value in r0 to D/A out
dec r0 subtract 1 from D/A value
bne loop
br start loop back to start
I was surprised to see that it took ~34 ms to run through all the numbers from
0-7777, that is about 34 Hz. The manual says the 'settling time' is 6
microseconds. Is this fast enough for audio?
How would you convert a modern audio file into 12 bit integers?
Doug
