Digital, but not by the numbers

[abraxalito]

Originally Posted by DF96

Thanks, I hadn't seen that. It appears to be a hybrid digital filter, using delay lines and multiple DACs to achieve an analogue implementation of a digital filter. Interesting idea, but I suspect a good implementation would require very good mixed-signal design techniques.

I concur so far.

The aim appears to be to compensate for the sinc frequency response of a simple NOS DAC. It is unclear to me what advantage this gives over a conventional oversampling DAC.

I can pontificate on that topic for a paragraph. A conventional oversampling DAC would need a digital filter prior, this would generate a result longer than 16bits and hence require dither followed by rounding which decreases the dynamic range, given the use of a 16bit wide path to the DAC chip. Using multiple DACs has an advantage I believe in terms of low-level distortion though this is still just a hypothesis, not yet tested by measurement. Multiple DACs also give more design flexibility because a greater output current is available. There's one further point that in this design each DAC is operating in RTZ (return to zero) which is a mode of operation I was curious to listen to.

I did indeed listen to a 2X OS filter into the same DAC chip (arranged in a quad, balanced configuration) and liked the sound less than this approach. But it wasn't a controlled experiment I admit.

Given that it greatly multiplies implementation problems while at the same time necessarily limiting the complexity of the digital filter design it seems to be on balance a worse way of doing it.

You're most welcome to your perception of it. Mine naturally enough differs from your own.

I would only be disappointed if I had been expecting something good. I didn't actually notice any ideas in the first post - the ideas, such as they are, appear in your blog.

Expectation bias doesn't only apply where the expectation is deemed 'good' by the expectant one.

[Pano]

OK. Is this a design philosophy thread, or a "let's design it and build it" thread?
Just so we know where it's going. Thanks.

[Trevor White]

Originally Posted by abraxalito

Wow, that's some compliment Trevor. Without even seeing the book I managed to transgress all the strictures? J.S. Bach eat your heart out.



EMC could indeed be a problem but as its not being put in the market, there's no need for regulatory compliance. As regards your other complaints, they're not an issue for the sound or the rudimentary measurements I've so far made.



Which measurements would those be?

You've got to be kidding me !!

You're still living back in the days of vaccuum tubes and point to point wiring and applying that to high speed digital design It's never going to work properly. And some measurements of the frequency spectrum would confirm that.

[abraxalito]

Originally Posted by Pano

OK. Is this a design philosophy thread, or a "let's design it and build it" thread?

Definitely design philosophy but with examples to illustrate that philosophy that the popcorn munchers can build for themselves if they so desire.

[abraxalito]

Originally Posted by Trevor White

You've got to be kidding me !!

How curious - there I was feeling quite convinced that the boot was entirely on the other foot.

You're still living back in the days of vaccuum tubes and point to point wiring and applying that to high speed digital design

Unlike you, I don't consider a clock rate of 2.8MHz 'high speed'.

It's never going to work properly. And some measurements of the frequency spectrum would confirm that.

The frequency spectrum of what though? When you say 'it' are you thinking that I'll build the whole system the same way I built the first prototype DAC? If so then you assume too much.

[fas42]

Originally Posted by abraxalito

I can pontificate on that topic for a paragraph. A conventional oversampling DAC would need a digital filter prior, this would generate a result longer than 16bits and hence require dither followed by rounding which decreases the dynamic range, given the use of a 16bit wide path to the DAC chip. Using multiple DACs has an advantage I believe in terms of low-level distortion though this is still just a hypothesis, not yet tested by measurement. Multiple DACs also give more design flexibility because a greater output current is available. There's one further point that in this design each DAC is operating in RTZ (return to zero) which is a mode of operation I was curious to listen to.

I did indeed listen to a 2X OS filter into the same DAC chip (arranged in a quad, balanced configuration) and liked the sound less than this approach. But it wasn't a controlled experiment I admit.

My take is that where abraxalito will "win" is in the minimising of fast rise circuitry in the circuit overall -- effectively making the digital work as much like low speed analogue circuitry as he can. It's the separation of high frequency, digital glitching from slow coach audio signals that confers most benefit in the "listenability" stakes ...

Frank

[abraxalito]

Originally Posted by fas42

My take is that where abraxalito will "win" is in the minimising of fast rise circuitry in the circuit overall -- effectively making the digital work as much like low speed analogue circuitry as he can.

Indeed - well spotted Frank. Treating the digital circuitry as just more analog circuitry is indeed part of the philosophy. So I steer away from the CMOS families which offer lashings of power supply glitches, use power supply voltages as low as I dare, attenuate digital waveforms with series resistors. Stuff like that.

It's the separation of high frequency, digital glitching from slow coach audio signals that confers most benefit in the "listenability" stakes ...

There speaks a man with broadly similar experience to my own in digital systems.

[Trevor White]

Originally Posted by fas42

My take is that where abraxalito will "win" is in the minimising of fast rise circuitry in the circuit overall -- effectively making the digital work as much like low speed analogue circuitry as he can. It's the separation of high frequency, digital glitching from slow coach audio signals that confers most benefit in the "listenability" stakes ...

Frank

digital design doesn't work well with slow rise time signals that you are advocating. Slow rise time signals cause oscillations and false triggering in logic devices. You can't treat digital processing like low speed analogue circuitry unless you want to go back in time to the UNIVAC which would never have been quick enough to process digital audio !!

[abraxalito]

You're speaking of too slow rise time signals, which can cause metastability in flip flops for example. However HC logic naturally generates slow rise time signals when fed with low supply voltages (I'm currently using 2.5V). No indication on the datasheet that any problems arise from using such a low supply - indeed the logic is specified down to 2V and will probably work even lower.

[Trevor White]

Originally Posted by abraxalito

How curious - there I was feeling quite convinced that the boot was entirely on the other foot.



Unlike you, I don't consider a clock rate of 2.8MHz 'high speed'.



The frequency spectrum of what though? When you say 'it' are you thinking that I'll build the whole system the same way I built the first prototype DAC? If so then you assume too much.

2.8MHz square waves with fast rise times which means at least a 28MHz bandwidth to maintain signal integrity !! Your rats nest circuit design topology will be ringing all over the shop not to mention crosstalk and false trigger !! It will certainly not be by the numbers !!