[sdiy] How to use an analog time-domain multiplexer?

Richard Wentk richard at wentk.com
Fri Jun 4 15:53:32 CEST 2021

If you’re sampling at 100k and max f is (say) 20k the filtering won’t need to be complicated.

You will need S&Hs, however.

Which is one of many reasons why this won’t work, because the hard part is building a super-low-distortion audio rate S&H with minimal jitter and no switching noise.

That’s not going to be clean and transparent unless you spend a lot of time and money on it.

And that still won’t help, because you’re effectively trying to do Pulse Amplitude Modulation at RF rates down unterminated plain wire. 

This guarantees all kinds of time smearing and reflections, so the odds are good that what comes out the other end will be a mess no matter how much filtering you do at each end.

You might just about get PAM working for a single audio channel with a sample rate of 44k, and if you try really really hard you can possibly push that to 100K or so for stereo. It probably won’t sound great, but it has the potential to more or less work. 

But any form of PAM at RF rates without RF connectors and RF design techniques is pure fantasy. And even with them, the cost, complication, and likely low fidelity make it unworkable. 

There are good reasons this kind of thing is usually done digitally.


> On 4 Jun 2021, at 09:14, Tom Wiltshire <tom at electricdruid.net> wrote:
> For the bandlimiting filter, I don’t think it matters so much. For the reconstruction filter, I’d go with multiple feedback filters because the stopband performance is better, and you might well need a filter than removes stuff a lot higher than the cutoff. Since they’re inverting, it would make sense to have the bandlimiting filter the same (two inversions in each channel, only need to work out one set of filter values).
> I recommend you build a simple prototype to see if your idea is feasible:
> 16 inputs, to sixteen filters to a 4067, then a short wire to a 4067 and sixteen more filters. Plus you need something to clock/address the 4067.
> Say each channel is sampled at 100KHz, so the switching rate is 1.6MHz. So the filters need to have a cutoff of 25KHz, but still be effective at 1.6MHz. That’s a tough spec.
> Also (unless we add sample-and-holds) we’re basically turning each channel off for 15/16ths of the time, so make-up gain is going to be required at the far end. That could be combined with the filter.
> ==================
>       Electric Druid
> Synth & Stompbox DIY
> ==================
>> On 3 Jun 2021, at 17:07, cheater cheater via Synth-diy <synth-diy at synth-diy.org> wrote:
>> No matter what I do I can't get digital as cheap as this. And at 100s
>> of ports per modular synth, that easily adds up. You'd have to have
>> one ADC and DAC pair cost less than $0.08. I don't think that's
>> feasible. And that's not counting all the digital glue, line drivers,
>> etc.
>> I'm looking at the Nexperia 74HC4067. It seems like it can be switched
>> at 8.33 MHz. Which would mean 520 kHz sampling frequency per channel.
>> Can someone confirm this?
>> Here's the datasheet:
>> https://www.mouser.at/datasheet/2/916/74HC_HCT4067-1597878.pdf
>> What sort of band limiting and reconstruction filters should I be
>> using? What topology / design?
>> Thanks
>> On Thu, Jun 3, 2021 at 3:13 PM Benjamin Tremblay <btremblay at me.com> wrote:
>>> I know the old copper telephone system used multiplexers like this for phone-grade audio. Bell Labs established the idea that you must have sample rate 2x the highest frequency to prevent hard distortion, and that applies to analog delay lines, too. So if you run the multiplexer so each channel gets 22kHz of air time, you will get distortion after 11KHz.
>>> Lesson learned: as long as you constrain your solution to a spec, you’re going to hit the target, and if not, roll back your spec.
>>> The phone system had the advantage of only having to support the original telephone spec. So quiet, you can hear a pin drop, through a carbon microphone.
>>> Personally I think you can get full audio spectrum, with some artifacts. The higher the clock frequency, the easier to hide the artifacts.
>>> But, let’s consider the possibility of using ADC/DAC and serial ports. Seems like a USB “sound card” (as the Asians call them) podcast mixer is competition for passing multiplexed audio over a single cable.
>>>> On Jun 3, 2021, at 7:03 AM, cheater cheater via Synth-diy <synth-diy at synth-diy.org> wrote:
>>>> Switches like the 4067 have the bandwidth to pull 16 audio channels
>>>> through one wire, but it's not entirely clear to me how to use them
>>>> correctly. I roughly know that I need to have a band limiting filter
>>>> before the analog switch, then i need a level shifter / line driver,
>>>> then the signal goes into a cable, then there's a receiving buffer /
>>>> level shifter, then the other switch, and then possibly an s&h and
>>>> reconstructing filter. However, I just don't know that much about how
>>>> this should be done.
>>>> For example, I don't really know much about the S&H / reconstructing
>>>> filter part. Isn't the signal coming out of the switch basically a
>>>> multiplication of two signals: (desired audio) * (gate), where the
>>>> gate happens at 1/16 the switching frequency. Looking at data sheets,
>>>> the Nexperia part seems to have a t_on and t_off of roughly 60 ms
>>>> each, meaning it can switch at ~ 8.33 MHz; so one channel would be
>>>> sampled at 520 kHz. Given that the channel only /contains/ 22 kHz
>>>> audio, wouldn't (desired audio) * (gate) only contain two frequency
>>>> bands? one's at 0-22 kHz (the desired audio) and an alias around 520
>>>> kHz. 520/22 = 23.6.., So wouldn't reconstruction be as simple as
>>>> putting down a simple 2nd-order filter at 25 kHz?
>>>> Of course you don't want to run at 8 MHz if not necessary, so maybe
>>>> the switching frequency could be lowered to 4 or 2 MHz, but a filter
>>>> could still easily separate the desired audio with the aliases.
>>>> Am I on the right track here?
>>>> Regarding line drivers... what sort of problems could I run into here?
>>>> I understand I'd need to look at the GBW of the op amp and make sure
>>>> it fits what I'm doing, but other than that?
>>>> I'd like to run this signal over a normal mini jack cable - so I'm not
>>>> sure if this requires any special massaging.
>>>> BTW, is it a good idea to run the analog switches using Gray code, or
>>>> is sequential selection better?
>>>> _______________________________________________
>>>> Synth-diy mailing list
>>>> Synth-diy at synth-diy.org
>>>> http://synth-diy.org/mailman/listinfo/synth-diy
>>>> Selling or trading? Use marketplace at synth-diy.org
>> _______________________________________________
>> Synth-diy mailing list
>> Synth-diy at synth-diy.org
>> http://synth-diy.org/mailman/listinfo/synth-diy
>> Selling or trading? Use marketplace at synth-diy.org
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at synth-diy.org
> http://synth-diy.org/mailman/listinfo/synth-diy
> Selling or trading? Use marketplace at synth-diy.org

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://synth-diy.org/pipermail/synth-diy/attachments/20210604/79b8a559/attachment.htm>

More information about the Synth-diy mailing list