NOS Dac filters

[studiostevus]

Quite some discussions in the past about this topic. From what I could read, most posters like the sound of a unfiltered NOS DAC (I am using TDA1541a), but liked roll-off compensation even more!

My question: Has anyone ever tried a low-pass notch filter on a NOS DAC? It seems that the frequency response could be tuned to both compensate the roll-off and the aliasing at 44.1khz...... not sure about the phase response though...

Capture.JPG

[oshifis]

What is the mathematical expression (in terms of poles and zeros) of the sinc(x) function?

[abraxalito]

I've tried peaky LC filters on my NOS DACs, haven't much liked the sound. Much prefer the solution I have now, but it'd be too expensive to implement on TDA1541...

Besides the sound (which perhaps others won't mind, its a personal thing), LC filters have tolerance and stability issues where high Q is being used to cancel out the roll-off.

[studiostevus]

Originally Posted by abraxalito

I've tried peaky LC filters on my NOS DACs, haven't much liked the sound. Much prefer the solution I have now, but it'd be too expensive to implement on TDA1541...

Besides the sound (which perhaps others won't mind, its a personal thing), LC filters have tolerance and stability issues where high Q is being used to cancel out the roll-off.

So what are you using?

[studiostevus]

Originally Posted by oshifis

What is the mathematical expression (in terms of poles and zeros) of the sinc(x) function?

I'm afraid you are asking the wrong person.... sinc(x) probably looks pretty terrible. That being said, Zanden uses multiple notch filters to filter aliases at n * 44.1khz.... looks horrible, sounds wonderful (so they say).

[abraxalito]

Ah, glad you asked that

I'm using what I call a longitudinal array of inexpensive DACs, otherwise known as a transversal filter. It means I can program the frequency response in as resistor values in the I/V stage. Have a look at my blog post on my 'MOS DAC prototype'. Questions welcome of course

[Ken Newton]

A low-pass notch filter will leave much of the ultrasonic image replications unfiltered.

[abraxalito]

Yeah, that was the same impression I got when I trawled through the Zanden patent, strangely enough. The notch is deepest in the middle (obviously) but that's precisely where the sin(x)/x curve is already doing its best work. A canyon is really called for, not a notch.

[oshifis]

NOS roll-off occurs at first-order sample and hold. The resulting staircase waveform is also sensitive to jitter. I have an idea to getting rid of both:
The analog pulses coming from the DAC could be passed through a C-R differentiating circuit that converts them to short spikes. The negative spikes should be removed by a diode. The peak amplitude of the positive spikes is identical to the original pulses, and their waveform is an exponential decay (following the e^(-RC/t) function). If the waveform of each pulse is identical, it will be insensitive to jitter. Also the short pulses are different from the S/H in that there is no sin(x)/x rolloff.
The pulses still need low-pass filtered to get smooth audio signal. I am not sure if this should be a brickwall filter, or a more relaxed one.
The pulse-count FM demodulator works on similar principle.
The disadvantage of this solution would be the low energy content of the pulses. An amplifier stage will be needed to amplify the raw pulses or the low-pass filtered signal.
Please comment this idea.

[abraxalito]

Why remove the negative spikes? Signals spend as much time moving in a negative direction as they do in a positive one. Removing the negative going impulses sounds like a recipe for serious distortion to me.

I think you omitted that with a differentiator the output is going to be dependent on the rate of change of the signal - how do we know that the DAC's output never gets slew-limited? Also the settling transient typically won't be monotonic. Using a differentiator will mean glitches get amplified more than the wanted signal.

Overall its sounding like your proposed medicine is worse than the disease its setting out to cure. I don't find jitter to be audibly objectionable on the output of my DACs.