adamdea Posted September 9, 2018 Share Posted September 9, 2018 4 hours ago, Miska said: This is article discussing the topic: https://www.cco.caltech.edu/~boyk/spectra/spectra.htm (the most valid for this is the section VI. Instruments without harmonics https://www.cco.caltech.edu/~boyk/spectra/11.htm#b Miska thanks for reminding us of that article. There are a couple of points which occur to me. - if we look for example at the jazz rimshot shown at figure 12 b) , am I reading this correctly: it appears to show a rise time of 148 us ie 6 and a bit sampling intervals at 44.1 kHz. That gives me the impression that the overall envelope could be captured in 44.1 kHz without too much problem. Certainly the envelope does not look like anything like a pulse. I guess there may still be enough of a discontinuity but how much ringing would that generate? Also at what level would it me reasonable to regard ringing as insignificant- could you resolve a tone at below ambient noise level if it only lasted for say 3 ms? I did play around a while ago with trying to create preringing in audacity by downsampling tracks but in the end when I proudly presented my homework to Jim lesurf he politely but firmly told me that I shouldn’t worry my pretty little head about it and that even the thing which looked like pre-ringing to me could not be shown to be. I can’t now remember what the reason was but I think it was to do with the very short duration and the potential for windowing artefacts etc. Another question is what percentage of sound over 20kHz (table 1) is enough to matter. The piano for example is shown to have some high frequency components it’s output. But with 0.02% over 20kHz it does make one wonder whether it really matters. Certainly the piano recordings I have looked at seem to have no sigificant stuff over 20 kHz. You are not a sound quality measurement device Link to comment
mansr Posted September 10, 2018 Share Posted September 10, 2018 39 minutes ago, adamdea said: Another question is what percentage of sound over 20kHz (table 1) is enough to matter. Wrong question. Content in the filter stopband does not cause "ringing." What matters is the intensity at the filter cut-off. Link to comment
adamdea Posted September 10, 2018 Share Posted September 10, 2018 3 hours ago, mansr said: Wrong question. Content in the filter stopband does not cause "ringing." What matters is the intensity at the filter cut-off. Yes I get that. For most instruments the intensity will tend to decline with frequency though, and one might be forgiven for the sin of lazily assuming that the proportion of energy over 20kHz will be a rough indicator of the intensity of energy at 22.05. But I can see that this may not hold true. That said It does establish a ceiling for the about of energy at nyquist You are not a sound quality measurement device Link to comment
Popular Post Miska Posted September 10, 2018 Popular Post Share Posted September 10, 2018 7 hours ago, adamdea said: - if we look for example at the jazz rimshot shown at figure 12 b) , am I reading this correctly: it appears to show a rise time of 148 us ie 6 and a bit sampling intervals at 44.1 kHz. That gives me the impression that the overall envelope could be captured in 44.1 kHz without too much problem. Certainly the envelope does not look like anything like a pulse. I guess there may still be enough of a discontinuity but how much ringing would that generate? Also at what level would it me reasonable to regard ringing as insignificant- could you resolve a tone at below ambient noise level if it only lasted for say 3 ms? I did play around a while ago with trying to create preringing in audacity by downsampling tracks but in the end when I proudly presented my homework to Jim lesurf he politely but firmly told me that I shouldn’t worry my pretty little head about it and that even the thing which looked like pre-ringing to me could not be shown to be. I can’t now remember what the reason was but I think it was to do with the very short duration and the potential for windowing artefacts etc. Another question is what percentage of sound over 20kHz (table 1) is enough to matter. The piano for example is shown to have some high frequency components it’s output. But with 0.02% over 20kHz it does make one wonder whether it really matters. Certainly the piano recordings I have looked at seem to have no sigificant stuff over 20 kHz. That is not exactly enough to capture exact the shape of the attack, which has more HF components. But overall, in my opinion, CD's steep cut-off at 22.05 kHz is a problem, too close to audio band. Hires fixes that, especially DSD and highest PCM sampling rates which allow high frequencies to naturally roll off, without need for artificial band-limiting. So I don't see much point in sticking with old RedBook format for new content. And for RedBook the problems can be partially fixed. Another aspect of using extremely steep filter is that lot of instruments have some amount of frequency modulation (vibrato). When a filter cut-off is within the vibrato's frequency window (or harmonics of) while not covering the entire frequency window. This causes the signal to be cut on-off during the vibrato, causing pulsating signal as the frequency flips between passband and stop-band of the filter. This is one reason I recommend that when designing room correction filters, people would avoid creating very steep (high-Q) notches in the filters because it sounds bad (with side effect of increased ringing of the high-Q filter). So instead I'd recommend to use 1/9th or 1/6th octave smoothing for the response before applying corrections. Analog filters like 2nd or 3rd order are gentle enough not to cause such problems, but can have other problems if not designed correctly, but such are not steep enough for RedBook... Teresa and Currawong 2 Signalyst - Developer of HQPlayer Pulse & Fidelity - Software Defined Amplifiers Link to comment
mansr Posted September 10, 2018 Share Posted September 10, 2018 7 hours ago, adamdea said: Yes I get that. For most instruments the intensity will tend to decline with frequency though, and one might be forgiven for the sin of lazily assuming that the proportion of energy over 20kHz will be a rough indicator of the intensity of energy at 22.05. But I can see that this may not hold true. That said It does establish a ceiling for the about of energy at nyquist It's still a very indirect measure. Moreover, the attack profile at that frequency also matters, not just the total energy, peak power, or other simple measure. Link to comment
Shadders Posted September 15, 2018 Share Posted September 15, 2018 Hi, Just to continue the analysis of transient noise in filters, comparing MQA Filter 14 to a Linear filter : The bandlimited pulse was passed through the filters, and the following time base waveforms obtained. They have been zoomed in to show the differences between the input pulse and output pulse. The difference between the input and output of the filters was calculated, where each output was shifted to align with the input. Given that the MQA has a phase delay, the time alignment was implemented to generate the result where the differences were smallest. As can be seen, the linear filter with a slow roll off adds noise, or distortion at the -120dB level, which as previously is a peak before and after the output pulse from the filter. The difference of the input signal pulse and output signal pulse is of the order of -300dB. For the MQA Filter 14, the difference between the input waveform and output waveform is worst case -30dB. The problem here is that the phase delay of the MQA filter results in the difference signal appearing worse than what it actually is. Therefore, the next step was to analyse the filter output spectrum, and compare to the input signal bandlimited spectrum. The sequence was to compute the fft of the signals, and then normalise the signal, since the linear phase filter had a larger attenuation compared to the MQA filter. The normalised fft of each output signal was subtracted from the input signal fft, to see the difference in the spectra. The results were as follows : As can be seen, both filters have an excess of difference which has the same envelope as the input pulse spectrum. Recall that the linear filter with 500taps has two image pulses before and after the main pulse on the output of the filter : The two pulses were removed from the time based signal and the fft, normalisation and difference to the input signal fft calculated, with the following results : As can be seen, the image pulses at the output of the filter add to the spectrum at the output as expected, and once removed, the difference spectrum is below -260dB. Therefore, the image pulses contribute to the extra spectra in the output of the filter, as added noise, whose envelope is extremely close to the input pulse. The MQA filter output has a very similar spectra in the same band as the input pulse, which is approximately 40dB less than the Linear filter. It is not possible to analyse this added noise in the MQA Filter 14 output in the time domain since the phase is non linear, and minimum phase impulse response. What can be stated is that the MQA Filter 14 also adds the noise which is due to the transient effect of the input signal. The linear filter was modified to increase the stopband attenuation to -300dB, and the transient related image pulses were NOT removed from the filter output, with the following difference spectrum compared to the input signal : As can be seen, the envelope of the input spectrum is completely removed, which means that under the transient condition, the image pulses are reduced significantly so as not to add to the output spectrum of the filter. Below is the difference on a log scale (dB) between the input pulse and the filter output, once the signals have been time aligned. As can be seen, the difference between the input pulse and output pulse is nigh on non-existent. That is, a linear filter with a sufficient stopband does NOT add any transient noise to the audio signal. There is NO smearing or blur (dispersion), and the output of the filter is an exact copy of the input signal. As long as the audio recording chain and playback chain use linear filters with a sufficient stopband attenuation, then any transient noise can be eliminated or significantly reduced (stopband dependent), and the blur (dispersion)/smear does not occur. Regards, Shadders. Link to comment
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now