Using two voices for 56K sample/second audio output rate -- OCS -- No CPU or Copper

mc6809e · 16 January 2012, 04:33

I can't verify this, but I think it's possible to output audio at a rate of 57K samples per second without using CPU interrupts or driving the audio channels with a copper list on OCS. DMA fetch/output is used.

Here's the proposed trick:

Set two voices on one channel (voices 1 and 2 for example) for an output rate of 28867 samples per second along with addresses of two audio buffers.

Next, start voice 1. Then, after 62 bus cycles, start voice 2. The samples of voice 1 will be interleaved with those of voice 2.

This is possible because the output of each voice is determined by Paula and not DMA. While DMA at the beginning of each scan line is used to fetch audio data, the output of the fetched values is timed by Paula.

This effectively doubles the output rate since we can cause with this interleaving of voices a change in audio output that occurs 57K times per second.

Finding the correct values to place in the buffers might be a challenge. Random noise passed through a high pass filter would be fine if a bright cymbal sound is desired, for example.

Another way is to simply take an original audio sample and place even samples in one buffer and odd samples in the other. This can be somewhat justified by considering what happens when performing a Fourier transform on real data.

h0ffman · 27 January 2012, 02:12

Ok, time for some thinking...

For example lets say the 57k sample data looks like this...

$00,$10,$20,$30,$40,$50,$60,$70

the interleave on it is

Chan1 $00,$20,$40,$60
Chan2 $10,$30,$50,$70

First idea i though would be to use the 2nd channel as delta, this then pushes the point to the next desired point inbetween the samples. For example $20 to $30 is $20 + £10, however that means that the Chan1 will keep going to the point..

However, if you delta Both channels, that might work???

Photon · 28 January 2012, 00:29

Bit depth and continuous playing (requiring AUDINTs) for longer than 2.27027 seconds of high quality sound is the problem, not frequency. Just test in Wavelab on PC and listen

16 January 2012, 04:33	#1
mc6809e Registered User Join Date: Jan 2012 Location: USA Posts: 372	Using two voices for 56K sample/second audio output rate -- OCS -- No CPU or Copper I can't verify this, but I think it's possible to output audio at a rate of 57K samples per second without using CPU interrupts or driving the audio channels with a copper list on OCS. DMA fetch/output is used. Here's the proposed trick: Set two voices on one channel (voices 1 and 2 for example) for an output rate of 28867 samples per second along with addresses of two audio buffers. Next, start voice 1. Then, after 62 bus cycles, start voice 2. The samples of voice 1 will be interleaved with those of voice 2. This is possible because the output of each voice is determined by Paula and not DMA. While DMA at the beginning of each scan line is used to fetch audio data, the output of the fetched values is timed by Paula. This effectively doubles the output rate since we can cause with this interleaving of voices a change in audio output that occurs 57K times per second. Finding the correct values to place in the buffers might be a challenge. Random noise passed through a high pass filter would be fine if a bright cymbal sound is desired, for example. Another way is to simply take an original audio sample and place even samples in one buffer and odd samples in the other. This can be somewhat justified by considering what happens when performing a Fourier transform on real data.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Kipper2K HDMI Audio Output	bbond007	support.Hardware	4	04 October 2012 00:50
Audio output caps (again)	jimbob	support.Hardware	10	02 August 2012 01:50
AHI uae.audio with up to 192KHz sample freq.	amiga	request.UAE Wishlist	4	18 March 2010 01:35
Paula output rate	snyp	Coders. General	5	18 February 2009 02:32
Q: recording 4-channels sound output and "Playback Rate"	jbl007	support.WinUAE	0	07 June 2005 18:23

27 January 2012, 02:12	#2
h0ffman Registered User Join Date: Aug 2008 Location: Salisbury Posts: 744	Ok, time for some thinking... For example lets say the 57k sample data looks like this... $00,$10,$20,$30,$40,$50,$60,$70 the interleave on it is Chan1 $00,$20,$40,$60 Chan2 $10,$30,$50,$70 First idea i though would be to use the 2nd channel as delta, this then pushes the point to the next desired point inbetween the samples. For example $20 to $30 is $20 + £10, however that means that the Chan1 will keep going to the point.. However, if you delta Both channels, that might work???

28 January 2012, 00:29	#3
Photon Moderator Join Date: Nov 2004 Location: Eksjö / Sweden Posts: 5,604	Bit depth and continuous playing (requiring AUDINTs) for longer than 2.27027 seconds of high quality sound is the problem, not frequency. Just test in Wavelab on PC and listen

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)