Pas de tout, Christophe, I would never dream to insult an honest Frenchman just beacuse his bride was not a virgin.
OK, if I am doing this right: with that rare species, a real 24 bit ADC...if we accommodate a 2V peak signal from the cartridge at 20kHz, a healthy but not outrageous overload based on a ref level of 70.7mV peak, and the ADC accepts +/-10V, we tolerate a flat gain of times 5; then at 20Hz we're working with our ADC LSB of about 1.2uV (20V/(2^24) ), and our reference level is ~707uV peak, which becomes 3.54mV after the gain. So we can digitize the reference level to a little better than 11 bits. If some of those 11 are marketing bits we're still worse off. Of course we can relax the overload at high frequencies and help things along.
Maybe a dodgy 11 bits is plenty? I still would prefer to do at least some shaping of the response ahead of the conversion. And of course the flat amp has to be exceedingly good if the cartridge noise is to be the limitation, but that is usually true anyway.
It's not surprising to me that people have touted the flat + DSP approach as being more accurate---but they know what they are listening to, and in general any change in audio evaluation is heard as preferred, unless it is terrible. And in some quarters the very notion of "digital" exerts a powerful bias, mostly instilling positive expectations. But it is not all that hard to get the analog domain transfer function correct. For either approach the cartridge loading is a large effect.
But to each his own.
??? No, SK is a conventional min phase filter, with normal group delay........
True as such, but there is a workaround by measuring at the exact same spot where the calibration file for the anechoic room was made.
Last edited:
That's disappointing. Kinda forces one towards DSP, doesnt it?
One of several sources mentions 3-4mS of group-delay or more is audible;
The book --> Communication Acoustics. An Introduction to Speech, Audio and Psychoacoustics.
And, while running thru the GD info on Internet, here is an interesting side issue revisited: View attachment DS668WP1.pdf
THx-RNMarsh
Last edited:
Last edited:
No. There are no wrong conclusions from the side of the author but I am not sure I understand the problem.
Is the following procedure correct?
Using basic test equipment, phase shift can be measured at spot frequencies and a curve plotted of phase versus frequency. The slope of the curve (which is actually group delay) is scaled off around sections of the curve and a new curve, of group delay versus frequency, is then plotted.
Is GD an independent attribute of a filter?
All filter types and categories have known, predictable calculable phase change with frequency. If GD is the slope of this phase change vs frequency change, then each filter's GD is known or am I off here?
But we are much concerned on the GD numbers due to an electrical HP filter at 20Hz
What about comparing with the:
GD imposed by our loudspeaker enclosure acting as an acoustic filter (see Van Dickason “Loudspeaker Design Cookbook”)
Phase change vs freq due to our loudspeaker/our room interaction at 20Hz or about?
I still can’t find research results for the audibility of GD at such low frequencies.
There is a suggestion from John L. Murphy to extrapolate existing research data and normalise it by the time period of the frequency of interest
Discussion of Group Delay in Loudspeakers
With simple music material, it sounds better. With complex music, no.
As soon as you do an FFT with high resolution, you will confirm the old wisdom
This is strange.
It needs a bit of further investigation. If it is a suspended sub chassis TT, there is some plausible explanation.
If you have experimented with TT mats (vibration measurements and listening test), you will have noticed that good vibration isolation results, good 12”diaphragm damping and music playback quality, conflict with each other.
Demian follow the link to the “Channel D Corp”. AES paper below
That's interesting.
I always thought that it is better (in terms of headroom) to do the HF roll off in analog before the AD but I hadn’t thought that the all-digital RIAA has this benefit:
Doing the deemphasis in the digital domain, we lose in dynamic range, (but) the disadvantage is balanced by the complementary enhancement of digital resolution, due to treble preemphasis (as treble is of higher amplitude), in the frequency range where human hearing is at its most sensitive.
http://www.channld.com/aes123.pdf
http://lcweb2.loc.gov/master/mbrs/recording_preservation/manuals/Tektronix%20Cookbook%20of%20Standard%20Audio%20Tests.pdf
George
JC needs a pure - not audible HP filter (switchable in-out) ... one which GD has no audible affects.
Not mentioning speakers here as this is not the speaker forum..... BUT, the abrupt phase transition at resonance... ported especially.... is a cause of much audible GD.... according to pschoacoustic research, the abrupt changes are more audible than smooth changes.
THx-RNMarsh
Group delay per se is harmless. Its variation wrt frequency isn't.
It is true that for simple frequency-dependent analog circuits, the highest GD value may be close or equal to the variation over a 0 to ∞ frequency range, but things can be quite different for complex digital filters
I have asked for a reference that 2.5mS group delay is required for a subsonic high pass filter. I also suggested that number could be valid for higher frequencies. I showed an example of what can be made with a Sallen-Key filter.
As a response, I got a dismissive pointer to the AES library (like everybody is subscribing to that...). Well, you already know what follows...
As of what you can "easily hear all by myself": unless you are a whale, elephant, giraffe, hippopotamus, or rhinoceros, there is absolute no reason in trusting that.
I was not in concern, you know, and anyway, I don't believe a virgin French bride exists.Pas de tout, Christophe, I would never dream to insult an honest Frenchman just beacuse his bride was not a virgin.
I remember that, at my own marriage, the mayor asked our son if he wanted to come near him to marry his parents. He refused, of course.
If you assume that peak at 20KHz are the same as at 1KHz you are right, but something feels wrong about that. I have energy levels vs frequency but not sure how that translates to modulation. But I think you have 20dB better than you thought, which would be 14 bits at 20Hz.
What I don't know is how to do a front end that clips cleanly and recovers, but that is a std instrumentation problem. Sure some of you guys already know the answers
No - at least not for minimum phase filters, which is the vast majority of analog filters. All such filters have the same phase-frequency relationship, and group delay, regardless of topology.
You are right on the money, gpapag. For min phase filters, freq response, phase response and group delay are all restatements of the same thing. Know one, know them all, for all filter topologies.gpapag said:
Yes, much talk on this thread of varying GD by selecting topology whilst maintaining a filter f response is meaningless with classic analog filters, unless one uses all pass filters or steps into the wonderful world of digital filters.
Waly, I'm not singling you out.... I do a long run of internet reading and then come back with info. Its either in the PDF's or in their references. You can search the AES without being a member and anyone can download for a fee. Even I, as a member, have to pay a fee.
So, you do/pay for your own searches, thank you.
The GD numbers seems to vary ... some of that is found to be because the audible amount varies with the listening SPL..... whether headphones (lower GD number)or speakers (higher GD) are used etc. Under varying conditions you could use 20ms. I choose to be conservative to cover all the possibilities and be darn sure no one can hear it.
THx-RNMarsh
A cite supporting your assertion that this was a published number might help. That's what you were asked for. So do you have one?
Real world cap tests -
There has been a lot of people measuring thd and fft of distortion of caps which shows virtually nothing there. What they did not show is the distortion with voltage dropped across them -- whether it is from LF junk in LP system repro or from filter action, EQ, etc.......
Here is some interesting (?) numbers for some of the DIY'er re polar and non-polar caps with voltage dropped across them; 10Hz and 2.5v rms:
All various brands from China and new -- except the 1997 NOS which is Sprague.
Ta electrolytic: 2.2MF/100vdc THD = .069%
Al electrolytic: 3.3MF/250vdc THD = .0086%
Al elctro NOS (1997) 5MF/250v THD = .0056%
Now, BP type:
10MF/50V THD = .02%
2.2MF/100V THD = .047%
4.7MF/50v THD = .037%
[Cap was in series, load varied to get 2.5v drop across cap.]
THx-RNMarsh
There has been a lot of people measuring thd and fft of distortion of caps which shows virtually nothing there. What they did not show is the distortion with voltage dropped across them -- whether it is from LF junk in LP system repro or from filter action, EQ, etc.......
Here is some interesting (?) numbers for some of the DIY'er re polar and non-polar caps with voltage dropped across them; 10Hz and 2.5v rms:
All various brands from China and new -- except the 1997 NOS which is Sprague.
Ta electrolytic: 2.2MF/100vdc THD = .069%
Al electrolytic: 3.3MF/250vdc THD = .0086%
Al elctro NOS (1997) 5MF/250v THD = .0056%
Now, BP type:
10MF/50V THD = .02%
2.2MF/100V THD = .047%
4.7MF/50v THD = .037%
[Cap was in series, load varied to get 2.5v drop across cap.]
THx-RNMarsh
My numbers suppose a 5mV rms reference level for 1kHz, 50mV rms for 20kHz, 500uV for 20Hz---just the way JA presumes, and upon which he bases his evaluation of overload margin, usually the input level for 1% distortion.If you assume that peak at 20KHz are the same as at 1KHz you are right, but something feels wrong about that. I have energy levels vs frequency but not sure how that translates to modulation. But I think you have 20dB better than you thought, which would be 14 bits at 20Hz.
What I don't know is how to do a front end that clips cleanly and recovers, but that is a std instrumentation problem. Sure some of you guys already know the answers
The 2V peak for 20kHz is a lot (based on the above, a 29dB overload), and isn't sustained in real-world situations; and indeed if you clip cleanly and recover quickly, of little consequence. But I'd rather not clip at that point at all.
In the design underway here, the first stage is a first-order lowpass at 2122Hz (the 75us tau). Then a shelf following accomplishes the correction to standard RIAA. This has the advantage of knocking down high frequencies early on, and provided there is enough gain in the first stage, the second stage contribution can be negligible. The availability of the lowpass, initially, is useful for other nefarious purposes yet to be disclosed.
John Roberts did a phono pre years ago with a somewhat similar approach, although he didn't have access to parts as good as they are now. He also priced it way too low and put it in a rather pedestrian enclosure, so regardless of how good it may have sounded it died aborning and left him permanently disgruntled---he's spent a good many keystrokes trying to discourage me from bothering
I have assured him that it is not intended to make any money, although sometimes things can surprise.
- Status
- Not open for further replies.