Indeed you do, if you simulate a design correctly then introduce 5, 10 and 20us of delay you can quickly see how 20 is awful, 10 is manageable and 5 is perfectly adequate when aligning drivers.
It obviously depends on the xover frequency as to how much of an effect this will have, but loosely speaking if perfection is your goal then you wont want to use anything other than 192.
As an example 2500Hz has a wavelength of 13.6cm. It takes 400us for sound to travel that distance and a delay of 200us will take you from perfect constructive interference to complete cancellation, 20us is only 10 steps from all to nothing = nowhere near good enough. 20 steps is enough to land you within the ball park, but 40 is needed to hit the target.
GREAT THREAD!
Fifthelement and thoriated, you and likely others express criticisms of active implementation which i have often considered, but my experience is that of a dabbler.
What exactly constitutes an acceptable Bit Error Rate could justify a thread of its own.
I listen via dvd player from source, or via 24bit 96 soundcard. Once i owned a 'better' 192 capable card. In home studio recording, i could subjectively detect the increased quality of a 192 dub, vs the 96 'copy', but it was subtle. With commercial recording it was less obvious.
I would LOVE to be able to fathom out how or what to do to get far lower BERs.
Are there any differential bit (gray code?) encoding methods for audio?
Also as a child of '78 i caught the drawn out death of vinyl. I have some nostalgia for it, but so many physical limitations as Coppertop rightly says. Passive crossovers are slightly more benign i feel. Passive components are well engineered generally, tried tested and all that. Their limitation of course is that if not carefully designed, phase and impedance can be crazy. Same to be said of analogue active, with extra slew rate issues etc. DSP trades some of those issues for others. Time convolution and pre-ringing are not things i want to add to any system, but particularly to a digital source where AA, BER and others are compounded.
My greater point:
It is generally understood that passive is limited, and what the limitations are. They can be mitigated to a point where theyre acceptable.
DSP in my opinion, has a looong way to go before it usurps a passive component, in terms of waveform distortion. Like H class amps... Had a BIG H class amp, sounded crass, so im loathe to try another switching amp again.
Fifthelement and thoriated, you and likely others express criticisms of active implementation which i have often considered, but my experience is that of a dabbler.
What exactly constitutes an acceptable Bit Error Rate could justify a thread of its own.
I listen via dvd player from source, or via 24bit 96 soundcard. Once i owned a 'better' 192 capable card. In home studio recording, i could subjectively detect the increased quality of a 192 dub, vs the 96 'copy', but it was subtle. With commercial recording it was less obvious.
I would LOVE to be able to fathom out how or what to do to get far lower BERs.
Are there any differential bit (gray code?) encoding methods for audio?
Also as a child of '78 i caught the drawn out death of vinyl. I have some nostalgia for it, but so many physical limitations as Coppertop rightly says. Passive crossovers are slightly more benign i feel. Passive components are well engineered generally, tried tested and all that. Their limitation of course is that if not carefully designed, phase and impedance can be crazy. Same to be said of analogue active, with extra slew rate issues etc. DSP trades some of those issues for others. Time convolution and pre-ringing are not things i want to add to any system, but particularly to a digital source where AA, BER and others are compounded.
My greater point:
It is generally understood that passive is limited, and what the limitations are. They can be mitigated to a point where theyre acceptable.
DSP in my opinion, has a looong way to go before it usurps a passive component, in terms of waveform distortion. Like H class amps... Had a BIG H class amp, sounded crass, so im loathe to try another switching amp again.
Last edited:
Again I lost my longish post. It's probably the internet server where I am dropping my login in the middle of my posting.
Again to sum it up briefly. I found out about this through some simulations I did to estimate how much accuracy I could obtain in VHDL DSP for LVDT position monitoring for an avionics application. The excitation is 2950 hz sinusoidal, and the desire was to exceed analog circuit accuracy in the digital domain.
I originally calculated using about 4-16x oversampling (about 32 real ADC samples per cycle with some subrate (to 4x) DSP filtering, or not) and got roughly 0.5-1.5% give or take error (amplitude variation) depending on the exact relative timing relationship between the sampled waveform and quantization. I eventually calculated that I should go as high as 200Ks/S or even higher to obtain the desired 0.1%. This is for a continuous 2950hz sine wave. (I do have some latitude to average a number of cycles or compensate for the timing variations dynamically if done properly to improve this rather than merely increase the sampling rate, but I imagine that such 'averaging' or dynamic timing compensation would be verboten even if possible for best quality audio).
My simulations showed this much error and the linear relationship between it and the sampling interval period. I went online and found out that it was no secret at all in the fields I mentioned in my previous post and that my numbers actually lined up well with those I saw online. So there it is.
Again to sum it up briefly. I found out about this through some simulations I did to estimate how much accuracy I could obtain in VHDL DSP for LVDT position monitoring for an avionics application. The excitation is 2950 hz sinusoidal, and the desire was to exceed analog circuit accuracy in the digital domain.
I originally calculated using about 4-16x oversampling (about 32 real ADC samples per cycle with some subrate (to 4x) DSP filtering, or not) and got roughly 0.5-1.5% give or take error (amplitude variation) depending on the exact relative timing relationship between the sampled waveform and quantization. I eventually calculated that I should go as high as 200Ks/S or even higher to obtain the desired 0.1%. This is for a continuous 2950hz sine wave. (I do have some latitude to average a number of cycles or compensate for the timing variations dynamically if done properly to improve this rather than merely increase the sampling rate, but I imagine that such 'averaging' or dynamic timing compensation would be verboten even if possible for best quality audio).
My simulations showed this much error and the linear relationship between it and the sampling interval period. I went online and found out that it was no secret at all in the fields I mentioned in my previous post and that my numbers actually lined up well with those I saw online. So there it is.
Last edited:
Again brilliant. I performed a transfer function and Bode analysis of an aileron levering actuator, using 2 LVDTs for a Uni assignment. One of my conclusions was that the ADC or matlabs samplerate was not sufficient, the highest excitation frequencies where poorly resolved in dV and dt. I had to conclude the TF was 1st order, since the high end was limited to a degree, that i never reached the 45° phase point (so it could conceivably be a dogleg 2nd order TF).
So... Going from 200ks/s at 16bit... 3.2MHz sample rate? Is that correct?
So... Going from 200ks/s at 16bit... 3.2MHz sample rate? Is that correct?
Last edited:
Hi, mondogenerator -
Good to hear from somebody with a similar experience. We never got anything like this from the 'perfect sound forever' crowd - For them it was all "Look! 0.003% distortion (0.1db below overranging) and it will do a full amplitude sine wave at 20Khz! What more do you want?" Er, how about good sound and even a speck of technical integrity and intellectual curiousity? Thought the Sony Advertising Pod People were going to take over audio for 15 years. Oh, they left their mark, alright.
Good to hear from somebody with a similar experience. We never got anything like this from the 'perfect sound forever' crowd - For them it was all "Look! 0.003% distortion (0.1db below overranging) and it will do a full amplitude sine wave at 20Khz! What more do you want?" Er, how about good sound and even a speck of technical integrity and intellectual curiousity? Thought the Sony Advertising Pod People were going to take over audio for 15 years. Oh, they left their mark, alright.
Last edited:
Fair enough!
May be one could use the oversampling mechanism to get shorter delays without going to massively higher SRs.
The XTA DP4xx series of speaker management systems for example uses a 96k SR but the delay is adjustable in 0.325µsec steps.
Unless there is another way of achieving this that I don't know anything about in my ignorance.
They may have a master clock available divided down for system functions that has that granularity, use both transitions of a clock in registering to provide those steps, or somewhat less likely, a PLL to multiply frequencies sufficiently to get those steps or even a tapped delay line.
this is where perhaps amplitude precision is less important than sample rate, or oversampling. I once had a technics MASH cd player. Got it dirt cheap and it eventually expired. Radial swingarm laser carriage, sounded great and i believe it used a low bitwidth DAC maybe 4bit, and a x16 oversample rate.
Yes, I think that is mostly the reason.
I have run into problems crossing circa 7KHz. At a 96KHz sampling rate, it's OK.
Moving the driver is best, if possible.
Yes, of course. That's where the 96K I mentioned came from.
"Mea rotarus non primum, est."
What I mean is you don't need a higher sample rate as such; merely (correctly) interpolate between your existing samples and you can still play back at 44.1 kHz with almost infinitesimally fine graduations of delay from the original.
I would imagine that one can also do that with an all-pass filter to a certain extent using an IIR -bi-quad. As it stands the digital delay function block built into the analogue devices software for the DSP chip that I'm using doesn't have any delay option other then delaying in discrete steps with the smallest being 1 sample period. I get the impression that interpolating and resampling would require considerably more processing power then the simple delay that it provides. Of course the DSP chip does have ASRCs already built in, but they aren't able to work in the way described. I can use FIR filters with the chip but I know almost nothing about calculating their parameters.
Delays in digital aren't restricted to integer multiples of the SR. Check out 'Fractional Delay Filters' if you'd like to learn more.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.