Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Brian Campbell]

I would not exactly call this steep.
If you extrapolate the curve it looks like a 12 dB/octave slope.
This means 40 kHz is only 18 dB down, this could be brought back with an equalizer.
It's exactly what Schoeps does with their "CMC 6 Ug xt" -type mics.
If you want to see a really steep slope look at the brick wall antialiasing filter of any ADC.
It goes down to at least -96 dB or even -140 dB within the range of, e.g., >20 kHz to 22 kHz (~1/10th octave).
This is about 1000 dB/octave!

My understanding is that modern delta sigma converters are oversampling in the Mhz range and allows for a much more gradual anti-aliasing filter roll-off.

[klaus]

Let's bring this discussion back down to a level that can be understood and appreciated by the majority of this forum's readers.

Please explain in simple terms what your previous post tries to convey.
KH

[Kai]

Please explain in simple terms what your previous post tries to convey.
KH

I'll do my very best to tell the story without using figures
Frequency response and timing is strictly connected, no matter what technology is used.
Be it analog or digital.
There is no way around this. Even the famous "phase linear filters" do smear a signal the same amount as any other, they've just got a different shape that might even sound worse (pre-echo).
The steeper a filter works the more it shifts timing.
An AD-converter needs to remove every signal above one half of its sampling rate.
Every signal above will mirror at this border into the audio range, causing clearly audible artifacts.
No matter how the AD-converter is built, conventional+analog filter or with oversampling+digital filtering, finally the exact same very steep filter needs to be there.
This filter effects signal timing up to one octave down from its cut off frequency.
On a CD this means the complete highest octave off the audible range.

In contrary every microphone has a high frequency roll off with a comparably very soft steepness.
Therefore every acoustic analog source contains a significant amount of ultrasonic signals and the time response of a microphone is much better then the one of an AD-converter.

One needs to use higher sampling rates to avoid time smearing in the audible range, and loss of ultrasonic content.

 Mission accomplished, no figures used in this explanation

[Brian Campbell]

Yes and the point of oversampling is that the anti aliasing filter on the analogue input can be pushed well out of the audible range. The result is then down sampled to the required sample rate. The decimation filter on the digital side is much easier to construct and does a much better job. As I said in my first post the sample rate discussions rage on. What prompted this discussion was a QC from a so called audiophile label that reject a mix on the basis that it had no ultra sonic content above 30khz and therefore cannot be a high resolution recording. What a load of rubbish!
I was certain that a U67 's frequency response in the 30khz range was very poor to non existent.

[Kai]

... oversampling ... down sampled to the required sample rate. The decimation filter on the digital side is much easier to construct and does a much better job.

There is absolutely no difference, you cannot circumwent physics (read above).
You have a benefit if you do intermediate processing, but that's another story.
Sample rate conversion, on the other hand, is tricky if you want to have it well done.
I prefer working on the target sampling rate.

What prompted this discussion was a QC from a so called audiophile label that reject a mix on the basis that it had no ultra sonic content above 30khz and therefore cannot be a high resolution recording. What a load of rubbish!
I was certain that a U67 's frequency response in the 30khz range was very poor to non

It depends on how you define "no content".
In fact it's really hard not to have any (let alone the self noise of the recording chain) if your system has the bandwidth to record it.

[Brian Campbell]

There is absolutely no difference, you cannot circumwent physics (read above).
You have a benefit if you do intermediate processing, but that's another story.
Sample rate conversion, on the other hand, is tricky if you want to have it well done.
I prefer working on the target sampling rate.

I believe that the ability to have a low order analogue filter before the ADC is considered an improvement.
Digital filters can have perfectly linear phase response and high order without significant noise problems. Anyway...

Has any one here actually measured the response of a U67 above 20khz. Just curious.

[Jim Williams]

Analog anti-aliasing filters do create phase shift, but the topology does affect results. Common "Butterworth" filters used in pro audio converters will offer maximally flat response, at a price. Run a square wave and you will see ripple on the top of the waveform.

Less common Bessel or Linear Phase filters have a less steep roll-off but offer a linear phase result. You won't see the ripple you see with the Butterworth slopes. The Butterworth slopes will also re-orient upper harmonics in time, placing them in an incorrect order creating another form of distortion.

When I did filter designs for Mytek I used a 65k hz Bessel 2 pole filter, that allowed the extra bandwidth of the higher sampling rates plus a benign filter without errors.

[Kai]

Bessel ... filters ... offer a linear phase result.
... When I did filter designs for Mytek ...

Sorry, but this is simply not true. Bessel filters, like most other, do have a 90° phase shift at the crossover frequency, approximating 180° in the stop band.
Only the transition is softer.
This is very basic physics and there is no way around it, except if you compensate the phase shift for example with an all-pass-filter or in the digital domain.
Apogee used all-passes in their famous anti-aliasing-filters for 16 bit converters in the early ages of digital.

[Jim Williams]

I did not claim a Bessel filter has no phase shift, it does. It's a linear phase curve. 10 degrees at 1k will result in 20 degrees at 2k hz not 21 or 19 degrees. Read up on the late Dean Jensen's AES papers on the subject, it's good reading.  He showed that a linear phase curve is benign, non-linear phase will smear harmonics, hence the concept of the Jensen linear phase transformer designs. Many other filter topologies produce a non-linear phase curve. It's easily seen on a graphic analyzer like the AP. Square waves show the errors on a scope.

All recorded audio has phase shifts, they come with a reduced bandwidth. That is why my analog signal path has a 2 hz to 200k hz bandwidth minimum, it allows no phase shift in the 20~20K audio band. That includes the current feedback mic preamps at 30 mHz, modified microphones with the electronics at 300k hz,  the analog console with current feedback suming at 30 mHz bandwidth, the power amp at 500k hz bandwidth, current feedback headphone amps at 50 mhz etc. Phase shift comes from the filters in the ADC which is also a Bessel design.

All pass filters also create phase shift. It's the basis of many designs including old phaser guitar pedals that used cascaded all pass filters to create that modulated sound. GIC filters were also popular years ago as the audio path did not pass through them directly. These days the filters in the ADC chipsets do the work for you, but DAC's still require a filter unless they run DSD. My personal favorites were the 65k hz 2 pole Bessel Sallen-Key filters built with current-feedback opamps, 400 MHz bandwidth and no phase shift other than that linear 45 degrees produced by the 2 pole response. Those were done for a Mark Levinson high end DAC project that was based on current feedback, including the current to voltage stage, I still have one of those here, it's excellent.

[klaus]

I understand how to reduce low-frequency phase shift in a mic amp- a pressure gradient capsule's natural low-frequency roll-off helps. But how do you avoid such phase shift in the high frequencies, especially in Braunmühl-Weber double backplate designs?

Take the U87: leave out the negative feedback that brings down the capsule's response to a -4dB level @16kHz. and you get an unusably bright mic. Even a gentler approach- shunting some of the high frequency energy to ground, still does not eliminate the phase aberrations altogether in the region above 8kHz.

I'd love to understand how that could be done in style, theoretically and practically.

[Kai]

... U87: leave out the negative feedback that brings down the capsule's response to a -4dB level @16kHz. and you get an unusably bright mic. Even a gentler ... still does not eliminate the phase aberrations ..

I'd love to understand how that could be done in style, theoretically and practically.

The beauty is, no need to worry. Phase shift is strictly coupled to frequency response. If you compensate the frequency responses non-linearities you automatically compensate the phase response to linear.
And with linear I mean ruler flat.
Or, at least the more the frequency response is compensated, the more linear the phase response will be.

[J.J. Blair]

I'd just like to remind everybody that fully extended frequency response is rarely enjoyable or musical.  As Klaus and I discussed years ago, there're a reason we don't record with measurement microphones. 

[Jim Williams]

That is a subjective belief as many high end classical and jazz recordings are done with higher sampling rates and greater than 20 k hz bandwidths. The other benefit is reduced phase shift. Reduced bandwidths, THD and colors may be applicable for pop/rock productions but an orchestra will sound truer to the source without those equipment limitations.

[klaus]

That is a subjective belief as many high end classical and jazz recordings are done with higher sampling rates and greater than 20 khz bandwidths. (...) Reduced bandwidths, THD and colors may be applicable for pop/rock productions but an orchestra will sound truer to the source without those equipment limitations.

Another subjective belief that also cannot be scientifically verified. How many atrocious sounding classical recordings I have (correct that- I no longer have) which were "captured" with Danish 'measurement' mics... What some may call "truer to the source" I'd call "truer to a cold fish past its expiration date".

We agree on the sampling rate issue. A recent 'Golden Ears' listening symposium I participated in had some disagreements on some recording format issues, but one clear, unified outcome: all else equal, the higher the sampling rate the less annoying the sound.

[Jim Williams]

B+K mics can produce emotionally pleasing recordings, if applied well. For doubting Thoms I suggest a spin of some of Todd Garfinkle's excellent CD's found on the MA label. Those used a matched pair of mics, a single set.
www.marecordings.com