Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Oversample to crazy rates digitally then drop the analog filter--valid approach? (Read 6066 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Oversample to crazy rates digitally then drop the analog filter--valid approach?

http://www.head-fi.org/t/784602/chord-mojo-the-official-thread-please-read-the-3rd-post/14250#post_12446978

Again from the Mojo thread, this time actually regarding Chord's approach.

It is apparently to oversample the digital input to crazy rates digitally but then to omit a proper analog filter?
"1. Mojo has a very simple analogue topology with a single stage analogue section. This keeps the analogue component count small, so improving transparency. But this means the OP from the noise shapers has to be low out of band noise, that is one reason why the noise shapers run at 104 MHz not the usual few MHz. It also means that the filtering has to be done within the digital domain."

I'm not sure here whether he means the mojo's DAC has no reconstruction filter or has a very simple one ("single stage"?).

Now suppose you oversampled by some crazy amount.  Intuitively it would seem to me that as the oversampling ratio approaches infinity, the magnitude of the possible step discontinuities between oversampled samples on a zeroth order hold would approach zero, so that a reconstruction filter becomes more and more unnecessary.

"2. You make the assumption that stop-band performance is not important, that a simple analogue filter is good enough. That is an assumption, my listening tests have revealed that even 120 dB rejection is not good enough - increasing it further gave sound quality benefits - and you can't possibly obtain greater than 120 dB stop-band from an analogue filter."

I don't know the reasoning behind this, but I was going to say, well come back to me when your headphone amp has over 120dB SNR...???

"3. The nature of the filtering has very important time domain effects. You can't reconstruct transients perfectly (look at Whittaker Shannon sampling theory) without using an FIR filter with a sinc response - an ideal sinc FIR filter will return the original un-sampled bandwidth limited signal completely perfectly. An IIR filter, or analogue filter can't reproduce the original exactly - there will always be time domain differences."

It does seem that he got the bare fact right here, a lowpass reconstruction filter will always shift the phase of the supersonic frequencies somewhat, if oversampling by a pedestrian 16x.

To summarize, it seems that other than point (2), his reasoning can actually be considered sound, assuming a certain unorthodox set of priorities? (i.e. eliminate even ultrasonic phase shift, effectively unlimited CPU power on tap?)

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #1 –
Quote from: Joe Bloggs on
I'm not sure here whether he means the mojo's DAC has no reconstruction filter or has a very simple one ("single stage"?).

Now suppose you oversampled by some crazy amount.  Intuitively it would seem to me that as the oversampling ratio approaches infinity, the magnitude of the possible step discontinuities between oversampled samples on a zeroth order hold would approach zero, so that a reconstruction filter becomes more and more unnecessary.

Please note that by reconstruction filter here I mean the analog lowpass filter between the final digital output and analog output.  Obviously the digital signal always passes through a digital reconstruction filter on the way to analog conversion.  That seems to be the point--to oversample so much digitally that after digital reconstruction analog reconstruction becomes unnecessary.

(digital and analog reconstruction filters, are they actually called different things?)

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #2 –
512 to 1024x oversampling is pretty common. Above that there is little point. At 1000x it is basically infinite anyway, so going higher makes no difference.

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #3 –
So an analog lowpass filter can indeed be omitted with no ill effect?  Would the noise introduced by directly converting the zero-order-held bitstream to analog be e.g. lower than -120dB as Chord requires?

And what do you make of claim (2) (the 120dB thing)?

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #4 –
Even when using 1024x oversampling, you'd better check the resulting spectrum, as you may run into problems with EMC. You may have to include an EMC-type filter to get rid of the switching noise, which is - again - an analog low-pass filter. Its corner frequency is well outside the audio band, however, so the effects on audio phase will be minuscule.

Effectively, through oversampling, the work of the reconstruction filter, as required by the sampling theory, gets divided between the digital and the analog side. Exactly where you put the dividing line is determined by the choice of oversampling factor.

Both the digital filter and the analog filter have imperfections, as there is no such thing as a sinc filter in practice (it would have to be of infinite length). There can only be approximations. Hence there are phase effects, but sane reasoning would conclude for most converters that they ought to be inaudible. If someone disagrees, he's got the obligation to provide some credible evidence, or at least some reasoning that doesn't depend on belief in magical properties of human hearing.

Similar things can be said about this 120dB thing.

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #5 –
There's again nothing new here, lots of confusion (also of terms) and unsupported claims. "my listening tests have revealed that" does not support someone's claims. What was tested? How was it tested? What listening test protocol was used? What were the results? ...

1st point confuses quantization noise (shaping) with imaging.
It doesn't help that you then speak of "no reconstruction filter", even if you later say you mean "no analog lowpass" which imho just adds to the confusion. But yeah, it definitely should also have an analog lowpass filter.

2nd point can be rejected simply because it is a nonsensical and unsupported claim.

3rd point is that non-linear phase filters (like min phase) have a non-linear phase response... of course, but this tautological point seems like a red herring when looking at the previous points. The person that made this point is probably very confused about what he/she's talking about.
A typical DAC does not use a min phase filter to oversample. It uses such an analog lowpass to attenuate noise and images after the signal's sampling rate has been increased by some factor, such that the influence of the phase shift of the lowpass filter will be minimal (inaudible) even at 20 kHz.
"I hear it when I see it."

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #6 –
Quote from: Joe Bloggs on
So an analog lowpass filter can indeed be omitted with no ill effect?  Would the noise introduced by directly converting the zero-order-held bitstream to analog be e.g. lower than -120dB as Chord requires?

And what do you make of claim (2) (the 120dB thing)?

Omitting a low pass filter on a DAC does not introduce in band noise,  so the noise floor will be unaffected.  The advantage over over sampling is that you can have a very flat pass band, good image rejection using simple filters, and can use noise shaping to push quantization noise below other noise sources in the DAC.  How much you need depends on the design of your DAC, and varies from about 8 to a few hundred.

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #7 –
Most DAC's these days are noise-shaped, oversampled PCM, with a 50kHz LP noise filter, give or take. So I'm not sure what the question is.

It seems that the OP does not grok the idea of noise shaping. 

There is a deck at www.aes.org/sections/pnw in the powerpoint section on this subject.  Unfortunately, there is no recording of the talk.
-----
J. D. (jj) Johnston

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #8 –
Quote from: Woodinville on
Most DAC's these days are noise-shaped, oversampled PCM, with a 50kHz LP noise filter, give or take.

That definitely explains why I was seeing a fitler begin at 50 KHz and a complete cutoff at around 80 KHz in my sound card.

I was generating MPX signals digitally using Stereo Tool and calibrating it for baseband FM transmitter use where subcarriers that make stereo sound and RDS text possible are in the ultrasonic range.

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #9 –
Quote from: Woodinville on
Most DAC's these days are noise-shaped, oversampled PCM, with a 50kHz LP noise filter, give or take. So I'm not sure what the question is.
I don't think that this filter is that low.

Quote from: Chibisteven on
That definitely explains why I was seeing a fitler begin at 50 KHz and a complete cutoff at around 80 KHz in my sound card.
No, that is not the filter Woodinville was talking about. You're talking about the digital interpolation/oversampling filter.
"I hear it when I see it."
 

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #10 –
I think it IS the filter I was talking about, Xnor, he's seeing the cutoff of the filter that blocks the noise resulting from noise shaping.

If his filter he's reporting is an antialaising filter, he's sampling at 192kHz or something like that.
-----
J. D. (jj) Johnston

Re: Oversample to crazy rates digitally then drop the analog filter--valid approach?

Reply #11 –
Quote from: Joe Bloggs on
http://www.head-fi.org/t/784602/chord-mojo-the-official-thread-please-read-the-3rd-post/14250#post_12446978

Again from the Mojo thread, this time actually regarding Chord's approach.
It is apparently to oversample the digital input to crazy rates digitally but then to omit a proper analog filter?
"1. Mojo has a very simple analogue topology with a single stage analogue section.

In fact a classic oversampled DAC has at least two low pass filters.

One is typically implemented in the digital domain with a corner frequency very close to the Nyquist frequency of the highest audio signal that is reproduced.  IOW this digital filter can commonly have a corner frequency from 22,050  to 96 KHz, plus minus a few KHz that depends on the philosophy of  some designer some place.

The other filter is typically implemented in the analog domain, and its corner frequency is based on a desire that most reasonable designers have to stop signals related to the oversampling process.  Typically this filter has a fairly gentle slope and a corner frequency around 240 KHz.

None of these frequencies are cast in cement or sacred or holy.

Usually, there's additional low pass filters  in the system related to the signal processing, power amplification, and loudspeakers.  The low pass filters that are  inherent in the design and operation of the human ear typically have an even lower corner frequency. 

Thus anybody who attempts to claim a dramatic sound quality improvement from the absence of a low pass filter or one that has a very high corner frequency is expecting people to ignore what is usually the strongest and lowest corner frequency of all,of the filters in the signal path, being the listener's ear.