A good multiplexer?

[jamos]

Can anyone point me to a patch that contains a good, reliable, fast-as-possible 8 or 16 channel interslot multiplexer? I am unable to make mine work reliably.

Thx x 10^6

[Tim Kleinert]

Many different ways to do multiplexers, all with different pros and cons in terms of DSP efficiency and features.

Do you need maximum throughput rate?

Do you need all channels available simultaneously at the output stage?

[jamos]

Excellent questions. Tim. I suppose on the face of it, throughput is most important.. but right now I'm looking for examples to pick apart to figure out why mine is not working.

This is for a rather strange application: a pantomorph. for this I need responsiveness, ie speed, with minimum DSP load. Actually I don't even need a multiplexer, I'm most interested in the demultiplexer; the multiplexer is inherent in the design.

Briefly: the pantomorph I am working on does not multiplex signals, but instead weights the signals using a matrix of control sequencers, which are controlled with a staircase waveform. These are then summed and placed on a bus. I need to demultiplex them using the same staircase. However, what I'm finding is that I cannot get the timing right: when I break the staircase down with an a/d and then use those bits to trigger S/H modules (or even T/H), I am capturing bad data: I cannot align the sample time properly with the multiplexed signal.

The full patch is complex and dependent on controller settings, which is why I am not posting it. I need to break the system down a bit to isolate the relevant components and then experiment with them. Perhaps then I will post an example. In the meantime, working examples would help a lot.

[Tim Kleinert]

If maximum throughput is the prime consideration, things become more klutzy because some useful modules cannot run at audio-rate (eg. the mux modules) and others require triggering, which basically halves the system frequency.
Having all channel outputs available simultaneously is the next challenge, because this requires some kind of S/H-scheme which has to be efficient enough.

Here's a 8 into 2 multiplexer (one signal and one pointer channel) which meets the criteria of maximum throughput (8 channels @ 12kHz), all channels available, and max DSP efficiency. It's built around an exponential counter circuit which generates 8 consecutive one-sample 12kHz triggers when connected to a AD module. This is used to multiplex and de-multiplex the signals as efficiently as possible.

If max throughput isn't important, things become easier. In the above case, the DIY expo counter circuit could be replaced by an 8-counter module triggered at 48kHz. The throughput will be halved however (6kHz). If you go down back to control rate, things again become a lot easier, because now you have the mux and demux modules at your disposal. Throughput for 8 channels would be 3kHz in this case.

The biggest Houdini-style multiplexing schtick I ever had to pull off was 64 audio sawtooth oscillators through 2 channels via linear inter-/extrapolation.

cheers

---

8xMuxer_12kHz_TK.pch2
Description:	Multiplexing 8 channels into 2, at maximum possible bandwidth (12kHz); all channel outputs available simultaneously.	Download
Filename:	8xMuxer_12kHz_TK.pch2
Filesize:	2.12 KB
Downloaded:	3773 Time(s)

---

[Tim Kleinert]

Addendum:
The above patch utilizes a neat property of the AD module, but thus is limited to a fixed number of 8 channels.

Below is a more generalized concept for maximum throughput/parallel outputs multiplexing, which can run any channel count you like, with the throughput rate changing accordingly (96kHz divided by channel count). This demonstration features 16 channels at 6 kHz, but adding or deleting channels is very easy. (The numbering scheme of the required switch and compare modules is straightforward. Adjust channel count in the multiplex counter and you're done.)

---

16x2_mux_6kHz_TK.pch2
Description:	Generalized concept of max throughput/parallel output multiplexing.	Download
Filename:	16x2_mux_6kHz_TK.pch2
Filesize:	3.96 KB
Downloaded:	3833 Time(s)

---

[jamos]

Thanks you master, this is the wisdom I was seeking. I will go to the wilderness to meditate a few days on this.

Seriously, thanks. I am going camping for a few days starting tonight, and I will start poring over these when I return.

[jamos]

Thanks a million, Tim. Your elegant and simple design beat the hell out of what I was doing. It works AND it uses less DSP. That staircase generator is a neat trick!

It works perfectly with the control-sequencer level multiplexing I am doing, but in the process it revealed a quirk of the control sequencer: the output lags the input by 8 clock cycles. See the attached example. You'd think that, with In 1 selected (what I labeled Channel 9) that slider 1 on the control sequencer would affect pitch.. but actually it is slider 9. Easy to correct for once understood. I do hope that the gate outputs stay in sync with the main output at this speed, because I am going to use them to switch in a few different signals.

Thanks again -

---

MuxTest.pch2
Description:	Demonstrates control sequencer lag	Download
Filename:	MuxTest.pch2
Filesize:	2.83 KB
Downloaded:	3903 Time(s)

---

[Tim Kleinert]

[jamos]

That makes sense. My multiplexer output is subject to the latency twice, once at the input, once at the output. OK, knowing that little fact I can compensate accordingly.

Are the FX buses subject to a similar latency?

[Tim Kleinert]