Anti-aliasing filter requirements

[darkfenriz]

Hi again
I have a simple theoretical question about anti-aliasing filers for 44100Hz A/D convertion.
I've made some calculations and for -3dB at 18kHz and -20dB at 22050Hz one needs a 5th order chebyshev or 12th order butterworth.
Preferably for -3dB at 19kHz and -25dB at 22050Hz there is a need for 7th order chebyshev or 20th order of butterworth. Is this correct? Of course it is hardly possible to build an analog filter of 4th order with good accuracy, not to mention higher orders, stability...
Is 44100Hz so much flawed? Do we really listen to "mirror image" frequencies?

[hagtech]

Yes, the brick filter requirements are very steep. That's one reason we don't use them anymore. Upsample! Or use a benign filter and live with the aliasing. Many people like the effect.

No, it is not that hard to make high order analog filters of high accuracy. Why would you say such a thing?

jh

[JohnW]

It’s long been a fatal mistake with digital filter designers to have only 6dB attenuation at 22.050KHz (half fs) this does not prevent foldback images of HF components back into the audio band.

The designers of the best sound digital filter (the PMD100) understood this, that’s why the PMD100 has the full -120dB attenuation by 22.050 KHz! Clever guys those old PM lot - show me one other digital filter that has full attenuation by fs/2! so simple - but so miss-understood!

[Werner]

Originally posted by darkfenriz

Is 44100Hz so much flawed? Do we really listen to "mirror image" frequencies?

I have no experience wit really good ADCs, only with the offerings from AKM which you find in your typical $200-400 recording interface.

Here I found indeed insufficient suppression of out of band components in the (oversampling!) AA filters, leading to easily measurable aliasing components popping up in the 10-22kHz band when using 44.1k.

A good test is the Cardas sweep LP, where you can see the cartridge's distortion components rise up to fs/2, and then fold back.

The in-built decimation filter of a package like Adobe Audition has far better performance, with zero foldback. That's why I now record at 88.2kHz, and then do the decimation for CD production in software.


44.1k flawed?

Yes.

Analogue AA filters are problematic.

IIR oversampling digital AA filters lead to in-band phase shifts (? I am no IIR hero).

FIR oversampling digital AA filters lead to pre-ringing (as opposed to the often misunderstood 'ringing' of proper Sinc-based oversampling reconstruction filters).
You can aleviate this AA ringing if you add a low-order minimum phase rolloff, starting at 18kHz or so. But this then reduces the bandwidth of the CD-medium to something less than 18kHz, and adds phase shift in the pass band. A bit like a half-decent cassette deck.

So yes, it should have been something like Fs = 60kHz, not 44.1kHz.

[peranders]

darkfenriz, what is your output really?

Making a good 20 kHz brickwall filter is more or less impossible. Why do you think we got oversampling?

My old Denon DCD-1500 has 2x oversampling and a 7th order LC filter at 35 kHz in a metal box. Not particulary good. My father's old DCD-1000 had a 5 th order LC-filter at 18-20 kHz, non-oversampled, did sound even worse and you didn't have to have golden ears for that.

[darkfenriz]

Thank you for input.
Honestly I'm thinkig much about Pavel Dudek's opinion about harmonic distortion masking flaws of a digital source. It makes sense to, HD 'aplifies' all the very 'natural' spectum harmonic components and thus masking bad products of aliasing. Trading one form of ditortion into the other? Maybe that's why class A 'high' distortion amplifier/preamplifiers are so preffered?

The in-built decimation filter of a package like Adobe Audition has far better performance, with zero foldback. That's why I now record at 88.2kHz, and then do the decimation for CD production in software.

I do the same and FFT filter for dead sureness of no HF components.

IIR oversampling digital AA filters lead to in-band phase shifts (? I am no IIR hero).

Neither am I, but I think they do.

No, it is not that hard to make high order analog filters of high accuracy. Why would you say such a thing?

You mean passive or active?

Thanks again and best regards