Chord M Scaler Opinions

[adamdea]

The problem with an M-scaler is that if you follow Watts' reasoning more taps is always better (and in the Keith Howard article he says that you still need more than the M-scaler). Well that came out 3 years ago and anyone who has followed Chord's product development will know that by now they could come out with a new product with longer filters. If they haven't done this recently I'm guessing it's down to the pandemic, becasue the rate they churned them out over the previous decade was impressive.
It's a great business model becasue you can always have more filter taps/ a longer sinc window

If you follow the logic, as opposed to the Kool aid, you are better off doing this in a computer of arbitrary power running software which can easily be updated. You can pay for this or get it for free. But paying thousands for an immediately obsolescent bit of hardware makes no sense. Howevr I see Grimm has got in on the act with a hugely expensive upsampler (4 x oversampling- imagine!)

 

[Suffolk Tony]

The problem with an M-scaler is that if you follow Watts' reasoning more taps is always better (and in the Keith Howard article he says that you still need more than the M-scaler). Well that came out 3 years ago and anyone who has followed Chord's product development will know that by now they could come out with a new product with longer filters. If they haven't done this recently I'm guessing it's down to the pandemic, becasue the rate they churned them out over the previous decade was impressive.
It's a great business model becasue you can always have more filter taps/ a longer sinc window

If you follow the logic, as opposed to the Kool aid, you are better off doing this in a computer of arbitrary power running software which can easily be updated. You can pay for this or get it for free. But paying thousands for an immediately obsolescent bit of hardware makes no sense. Howevr I see Grimm has got in on the act with a hugely expensive upsampler (4 x oversampling- imagine!)

Us sensible, normal folk will borrow these very expensive bits of kit & try them extensively before making up our own minds.

I owned a Chord QBD76 for many years, and was assured that the new Chord Hugo, having lots more taps, must sound better. It didn't in my system, which is all I'm ever interested in. It took a DAVE to top the QBD. I've a couple of Pis knocking about, but try as I might I'm unable to conjure up the magic touch that renders them anything more than a very poor second to the DAVE/ MScaler.

 

[monya]

I guess Chord could have shoved the M scaler innards into the Dave and save folk a chunk of dosh although their marketing dept probably wouldn’t agree.

 

[mansr]

The problem with an M-scaler is that if you follow Watts' reasoning more taps is always better

And that's just plain silly. Here's an example of what is possible to achieve using a few hundred taps (395 in this case):

More examples here: https://www.audiosciencereview.com/...-rme-adi-2-dac-fs-tap-count.22124/post-734918

 

[DaveMc64]

I thought Watts' argument was that it improved matters in the time domain rather than the frequency domain which, as mansr shows above, needs nowhere near that number of taps.

I don't have a dog in this fight either way though I do use software upsampling myself.

 

[camverton]

The problem with an M-scaler is that if you follow Watts' reasoning more taps is always better (and in the Keith Howard article he says that you still need more than the M-scaler). Well that came out 3 years ago and anyone who has followed Chord's product development will know that by now they could come out with a new product with longer filters. If they haven't done this recently I'm guessing it's down to the pandemic, becasue the rate they churned them out over the previous decade was impressive.
It's a great business model becasue you can always have more filter taps/ a longer sinc window

If you follow the logic, as opposed to the Kool aid, you are better off doing this in a computer of arbitrary power running software which can easily be updated. You can pay for this or get it for free. But paying thousands for an immediately obsolescent bit of hardware makes no sense. Howevr I see Grimm has got in on the act with a hugely expensive upsampler (4 x oversampling- imagine!)

No, if they bring out a product with more taps the current m-scaler will still have the same beneficial effect on the sound in my system.

Using software is certainly an alternative if you feel the quality is as good and, most importantly, it works with ones chosen software. Upsampling in JRiver does not sound the same as HQ Player but JRiver, with Remote has a near ideal user interface for me whilst as far as I could find out HQ Player does not. For the OP this is largely irrelavant as he prefers to use a CD player.

I guess that your use of the term "Kool aid" says much about your bias on the subject! What was your impression of the m-scaler, which Chord DAC or other make did you use with it, and did you feel it would benefit from more taps?

Going on to the very interesting Keith Howard article he does point out that the Watt’s WTA (Watts Time Alignment) windowing algorithm is a closely guarded secret. This makes it difficult for theorists to come to a valid, substantive conclusion, and why their views may be useful as a caution but should not preclude an audition.

 

[mansr]

I thought Watts' argument was that it improved matters in the time domain rather than the frequency domain which, as mansr shows above, needs nowhere near that number of taps.

The time and frequency domains are equivalent through the Fourier transform. What's better in one is better in the other too.

 

[adamdea]

No, if they bring out a product with more taps the current m-scaler will still have the same beneficial effect on the sound in my system.

Can you explain this?
I assume you can't be simply making the trite observation that the m-scaler won't work any worse becasue something else works better.
The point is that the mscaler only makes sense if it applies a more accurate (in Watts' terms) filter than is already available in one of his dacs.

Going on to the very interesting Keith Howard article he does point out that the Watt’s WTA (Watts Time Alignment) windowing algorithm is a closely guarded secret. This makes it difficult for theorists to come to a valid, substantive conclusion, and why their views may be useful as a caution but should not preclude an audition.

Watts makes a big deal about his windowing function. But it can't be a function unknown to maths. This sort of thing is the Kool aid I referred to.

 

[adamdea]

And that's just plain silly. Here's an example of what is possible to achieve using a few hundred taps (394 in this case):


More examples here: https://www.audiosciencereview.com/...-rme-adi-2-dac-fs-tap-count.22124/post-734918

I note that in the Howard article he says:

"Fig 5 emphasises an important point: that the envelope of the sinc(x) function – which is finite valued for values of x from minus infinity to plus infinity – decays slowly with time. At 150 sampling intervals from its central peak the envelope has only decayed by a little over 50dB. The obvious question is: by how much must it decay for its contribution to inter-sample wave shape to become insignificant? That’s not a straightforward question to answer but if we say 100dB, to take the envelope amplitude below the 16-bit noise floor for a 0dBFS (full scale) sample, we can easily calculate what excerpt of the sync function is required. The envelope of the sync function is determined solely by the denominator f sin(x)/x, so it behaves as 1/x where x = Nπ(angle in radians), N here being the number of sampling intervals away from the central peak. For the envelope to be 100dB down from its peak value 1/x = 0.000001, which is equivalent to N = 31,831. This sample length is required either side of the central peak, so the total length of the sinc(x) excerpt is double this. In other words, for 44.1kHz sampling rate the total length of the required sinc function excerpt is 1.443 secs – pretty close to the filter length provided by Chord’s M Scaler."
Now one could be forgiven for thinking on reading this that one needs a great long filter in order to reproduce the signal to be accurate down to the -100dB level.But of course this is not true.

 

[mansr]

I note that in the Howard article he says:

"Fig 5 emphasises an important point: that the envelope of the sinc(x) function – which is finite valued for values of x from minus infinity to plus infinity – decays slowly with time. At 150 sampling intervals from its central peak the envelope has only decayed by a little over 50dB. The obvious question is: by how much must it decay for its contribution to inter-sample wave shape to become insignificant? That’s not a straightforward question to answer but if we say 100dB, to take the envelope amplitude below the 16-bit noise floor for a 0dBFS (full scale) sample, we can easily calculate what excerpt of the sync function is required. The envelope of the sync function is determined solely by the denominator f sin(x)/x, so it behaves as 1/x where x = Nπ(angle in radians), N here being the number of sampling intervals away from the central peak. For the envelope to be 100dB down from its peak value 1/x = 0.000001, which is equivalent to N = 31,831. This sample length is required either side of the central peak, so the total length of the sinc(x) excerpt is double this. In other words, for 44.1kHz sampling rate the total length of the required sinc function excerpt is 1.443 secs – pretty close to the filter length provided by Chord’s M Scaler."
Now one could be forgiven for thinking on reading this that one needs a great long filter in order to reproduce the signal to be accurate down to the -100dB level.But of course this is not true.

That calculation is only meaningful if using a naive truncated sinc filter. Nobody does that.

 

[tuga]

Gents, don't upset the fanboys.

 

[DaveMc64]

What's better in one is better in the other too.

While I agree the two are linked through the Fourier transform, I wouldn't agree that what is better in one is automatically better in the other.

As a trivial example, (and I'm delving back into my prehistory here) comparing a Bessel and a Butterworth filter, the Butterworth has a steeper cut off than the Bessel filter and might be considered better in that regard in the frequency domain.
However, the Butterworth has a more variable group delay than the Bessel (maximally flat group delay) so the Bessel would probably be considered better in the time domain.

 

[camverton]

Gents, don't upset the fanboys.

Ric, I expected better from you!

Using something because it makes an improvement in sound quality, as iirc HQ Player does for you, doesn't make one a fanboy .

As for being upset , don't be silly. Those of us who have found a benefit to using an m-scaler are the fortunate ones, as presumably are you with your chosen method. I'm not quite sure how to describe those who prognosticate without having the good sense to actually try one and at least see if their ideas hold up or need to be reevaluated.

In the end it comes down to how far, in the pursuit of accuracy, one needs to go and I suspect this will vary from one individual to another, hence the advice the OP is being given to try and then decide for himself.

 

[adamdea]

That calculation is only meaningful if using a naive truncated sinc filter. Nobody does that.

...and it is particularly ironic when applied to someone who tells you he is using a special secret window function.

 

[AndyU]

Ric, I expected better from you!

Using something because it makes an improvement in sound quality, as iirc HQ Player does for you, doesn't make one a fanboy .

As for being upset , don't be silly. Those of us who have found a benefit to using an m-scaler are the fortunate ones, as presumably are you with your chosen method. I'm not quite sure how to describe those who prognosticate without having the good sense to actually try one and at least see if their ideas hold up or need to be reevaluated.

In the end it comes down to how far, in the pursuit of accuracy, one needs to go and I suspect this will vary from one individual to another, hence the advice the OP is being given to try and then decide for himself.

Absolutely. Moreover I find it hard to understand why those who claim that you can better the performance of an MScaler with a raspberry pi and some arithmetic don’t just do so, market their box, make a fortune, spend the money on hookers and drugs and use what is left to promote world peace and universal health care.

 

[mansr]

While I agree the two are linked through the Fourier transform, I wouldn't agree that what is better in one is automatically better in the other.

As a trivial example, (and I'm delving back into my prehistory here) comparing a Bessel and a Butterworth filter, the Butterworth has a steeper cut off than the Bessel filter and might be considered better in that regard in the frequency domain.
However, the Butterworth has a more variable group delay than the Bessel (maximally flat group delay) so the Bessel would probably be considered better in the time domain.

Phase response is part of the frequency domain, something you have conveniently ignored.

 

[adamdea]

While I agree the two are linked through the Fourier transform, I wouldn't agree that what is better in one is automatically better in the other.

As a trivial example, (and I'm delving back into my prehistory here) comparing a Bessel and a Butterworth filter, the Butterworth has a steeper cut off than the Bessel filter and might be considered better in that regard in the frequency domain.
However, the Butterworth has a more variable group delay than the Bessel (maximally flat group delay) so the Bessel would probably be considered better in the time domain.

Frequency domain [in the fourier transform sense] = frequency and phase.
The Butterworth filter is *not* better in the frequency domain because of the group delay (ie phase) problem.
I remember one horseshit audio manufacturer article which tried to say that the same piece of music played backwards was the same in the frequency domain, hence {insert bullshit marketign nonsense}

 

[adamdea]

Absolutely. Moreover I find it hard to understand why those who claim that you can better the performance of an MScaler with a raspberry pi and some arithmetic don’t just do so, market their box, make a fortune, spend the money on hookers and drugs and use what is left to promote world peace and universal health care.

you can't better the mscaler's performance in the domain which counts (marketing). Incidentally I think @mansr has done so in the boring maths domain using Sox (which is sadly bound to be free). Have I remembered this right @mansr?

 

[Cesare]

Phase response is part of the frequency domain, something you have conveniently ignored.

I think you are missing the point. I think what he's saying is that if you look at a plot of the frequency bin magnitudes as 'frequency domain' then the plot may look to be better for one filter implementation, but it might have delay implications which are best explored in the time domain, or by looking at the phase of the bins (which are typically ignored when plotted as a graph).

Of course what you should really do is plot the z transform and talk about it's properties, but the number of people who would then understand what you are on about drops pretty quickly

 

[mansr]

you can't better the mscaler's performance in the domain which counts (marketing). Incidentally I think @mansr has done so in the boring maths domain using Sox (which is sadly bound to be free). Have I remembered this right @mansr?

The SoX resampler is indeed one of the better performing ones available. I didn't write it, though.