John Curl's Blowtorch preamplifier part II

Status
Not open for further replies.
Disabled Account
Joined 2012
These graphs show only internal (to the DAC) oversampling digital filter stop-band performance in digital domain. And only up to the 4x Fs (!) They do not show what happens at 8xFs or what comes out of D/A portion of the chips.
The real situation is more like this. First image is manipulated by me, the second one is from the last page of the PCM1794 data sheet. Note the signal amplitude at 8x Fs in the first one and raising noise floor in the second that looks just like in Richard's measurements.

yes. This is the source of my HF noise on the analog outputs. Obviously without any further filtering after the DAC.
.

-RNM
 
Last edited:
Disabled Account
Joined 2012
The real situation is that you also have an analog filter after the DAC. See the datasheets for a typical implementation.

At 8xFs, even a mild analog filter will counteract the rising HF noise.

edit: and if you're worried about the -110dB noise floor at 100Khz, it's easy enough to implement a more stringent lpf.

The real situation is -- what I showed; There is not an analog filter on this CD player output. Probably not in a lot of others as well of mid price or lower.

The noise is well above the noise floor and would be audible if the HF was LF. This particular one was at -60dB. That modest level is further amplified by next analog amp. Many VFA will not be very linear at HF freqs and IM etal will be produced. The HF level to the tweeter doesn't do the tweeter any favors, either.

I designed this little multi-stage LCR filter with parts on hand and put it on analog output.... the Hf is gone and the square wave is clean and flat topped with no ringing. And, it sounded better as well. Of course an even better filter could be made but for the test, this worked fine.

Dig filter.JPG


If mfr leave this extra filtering off, probably to save money, it is easy to put one in your player.



-RNM
 
Last edited:
Disabled Account
Joined 2012
Some background info if you find the ripple and HF noise on your analog output:

https://www.maximintegrated.com/en/app-notes/index.mvp/id/3853

Archimago's Musings: MEASUREMENTS: Digital Filters and Impulse Response... (TEAC UD-501)

BTW - If your output (reconstruction) filter is wrong in phase and ampl,, it will show on the reconstructed square waveform...... look for a nice sharp corners and flat top on sq wave.... just like the input waveform.

If you see ringing on an impulse or square wave in a review, you should be suspect that the output filter is missing or inadequate. This is done to save money and the THD within the BP of 20-20KHz will still look fine.


THx-RNMarsh
 
Last edited:
But I agree that digital output, especially close to Fs/2, is most usually not nice. This is a comparison of 21kHz sine from cheap generator GAG-810 with DAC sine 21kHz at 44.1kHz/24bit sampling. There are artifacts of 8-bit ADC fast converter used to capture data, but even though we can see how poor the "audio" DAC output is.
 

Attachments

  • gen 21kHz.png
    gen 21kHz.png
    41.3 KB · Views: 81
  • dig 21kHz.png
    dig 21kHz.png
    39.8 KB · Views: 75
Back to the basics: http://www.ti.com/lit/an/sbaa055/sbaa055.pdf

If a manufacturer doesn't do his job, you have everything in there to design a proper lpf, once you know the noise shaping profile of your DAC and the performance of its digital filter (both should be in the datasheet).

No. You will never be able to get rid of aliases close to Fs/2 (I have shown it at 23kHz approx.) with analog filter only. Or if yes, then the phase response would be horrible. The only improvement with analog LPF will be in reduction of ultrasonics HF rubbish.

Test your player or DAC with 21kHz and you will see that most of them produce aliases.
 
..... look for a nice sharp corners and flat top on sq wave.... just like the input waveform.

If you see ringing on an impulse or square wave in a review, you should be suspect that the output filter is missing or inadequate.

The Gibb's "ringing" is the correct output. You can not reproduce a 10kHz square wave with "nice sharp corners and flat top" from a CD. Unless that's not what you mean.
 
It is possible to "reproduce" square from CD without overhoot and ringing, but only at expense of mediocre amplitude and phase response. Such digitall filter would create lot of audible rubbish deep in audio band.

But who will get "square" through A/D process? Nonsense again, and bigger one. Everything above Fs/2 is to be cut away with very high rejection, thus steep filter.
 
It is possible to "reproduce" square from CD without overhoot and ringing, but only at expense of mediocre amplitude and phase response. Such digitall filter would create lot of audible rubbish deep in audio band.

But who will get "square" through A/D process? Nonsense again, and bigger one. Everything above Fs/2 is to be cut away with very high rejection, thus steep filter.

Pavel,

I think you need to expand on your response. No argument about trying to encode a high frequency band edge square wave. But in reproducing one it certainly is possible to build a filter that will tilt things into a square wave, but that is of course not an optimal approach.

I thought the issue RNM had raised is that the output of some commercial CD players are filtered for half the sampling frequency, but the filters let much higher frequency garbage through at levels that will cause problems for some of the following gear.

Now a Baxandall high cut filter would actually clean up much of this, but there are better approaches.

To me this means where a preamp is not required to increase gain, it may be desired for control and supplemental filtering.
 
I thought the issue RNM had raised is that the output of some commercial CD players are filtered for half the sampling frequency, but the filters let much higher frequency garbage through at levels that will cause problems for some of the following gear.

If by "levels that will cause problems" you mean -60dB (that's the latest figure quoted by RNM, this thread is a nightmare when you search information), it's good to keep in mind that the industry doesn't seem to think it is excessive. Quoting from this paper

Out-of-band noise (OBN) is troublesome in analog circuits that process the output of a noise-shaping audio DAC. It causes slewing in amplifiers and aliasing in sampling circuits like ADCs and class-D amplifiers. Non-linearity in these circuits also causes cross-modulation of the OBN into the audio band. These mechanisms lead to a higher noise level and more distortion in the audio band. OBN also leads to interference in the LF and MF band, compromising e.g. AM radio reception. To sufficiently avoid these problems, it is desired to reduce OBN power to below -60dBFS.

I'll let you judge if -60dB is or not a good line to draw.
 
If by "levels that will cause problems" you mean -60dB (that's the latest figure quoted by RNM, this thread is a nightmare when you search information), it's good to keep in mind that the industry doesn't seem to think it is excessive. Quoting from this paper



I'll let you judge if -60dB is or not a good line to draw.

Your cite just seems to pick that number, must have missed why.

If it costs me under a dollar to implement a switchable filter.... I think it isn't a bad idea to play with.

Now my listening space has a noise floor of 10 dBa and I certainly can hit peaks of 90 dBa with a headroom of another 20, so birdies at -60 dB FS just might be an issue.
 
I agree with everything wrote about the problems with signal frequencies closes to the half of the sampling one. (Aliasing)
On my side I was fighting, at the beginning of digital era, when everybody asked more bits, to make understood to my colleagues that higher sampling rate was far more important.

But now, We all use 96KHz as a minimal during recording processes. And I doubt any mike can produce any high energy signal at 40KHz !
And even, with 48KHz that was the standard for professional digital recordings at the beginnings, instant comparing the analog source and the digital proceeded one, it was difficult to hear a lot of differences. At this time, we made a lot of blind tests with a lot of people in our studios, to figure out this question.

Anyway, the artifacts created during the AD conversion will be there for ever. Not to confuse with the DA ones, that Richard is trying to eradicate, even if a low pass filter will help on both ones.

Again, in this matter, we cannot have a definitive answer by pure measurements. As we have to make a compromise between HF noise and distortions products, linearity and phase in the audio bandwidth, only our ears can tell us where is this best compromise. (A personal opinion ;-)
 
Last edited:
Disabled Account
Joined 2012
The Gibb's "ringing" is the correct output. You can not reproduce a 10kHz square wave with "nice sharp corners and flat top" from a CD. Unless that's not what you mean.

No its not what I mean... I mean you want the same signal waveform at the analog output of the CD player as the input. You get that After further filtering of the 'ringing'. This I have shown/demonstrated. Further elaborated with web site ref.

I also showed that you can filter the input to the ADC and get same results as filtered on DAC output.... a nice clean sq wave result. Such filter after the DAC output is also known as a reconstruction filter.

Note the scope photographs are the result from such LCR reconstruction filter and are in fact quit sharp and square now at the analog output. No rounded corners to imply rolled off high freq.

-RNM
 
Last edited:
Status
Not open for further replies.