In recent years 192kHz sampling frequency was announced to be the standard for modern audio recording. It appears to have significant advantages over 96kHz, but is it worth a try?

Do we really need 192kHz in modern audio recording?

General advantage of higher sampling frequencies

From the point of view of an engineer doing signal processing, there is no doubt in any way: Measurement systems in the field of data processing and acquisition do perform oversampling at about 100-1000 times of the desired highest frequency to get all of the available information correctly, since Nyquist's theorem required an ideal anti aliasing filter which does not exist in reality. So the idea, "the higher the sampling frequency the better for quality" absolutely is right. Read an article about a comparison of different sample frequencies a more detailed view on this issue: The advantage of 96kHz.

Higher sampling frequencies in audio processing

From the point of sound recording it is a bit different, since there is not a big part of energy in the upper bands so a slight linear distortion from a "normal" AA-Filter (having it's edge frequency at around 15kHz to provide a good stop band rejection) does not so much "harm" to the signal ,as most people think. In fact, the "errors" are mostly smaller than those introduced by our monitors.

96kHz had one big advantage: You can record in 96kHz and more or less down sample to 48kHz and resample to 44,1kHz without the fear of significant quality loss. Changing 48kHz recordings to 44,1 is not nice to do!  So the improvement from 48 to 96 years ago was THE big point (at least regarding this particular aspect, because on the other hand, the total improvement when comparing recordings of 44,1 and 96 directly was only little - even when having a full 96kHz system. And it is important to use a full 96kHz path with optimized anti aliasing filters at both ends of the recording chain: This refers to ADC and DAC and the whole digital path in between. The analog path has to be optimized that way, that filter edge frequency has to be far outside of the audible band.

Is 192 kHz "too high"?

So, 192kHz is even less important! As soon as you record through an AA Filter with an edge frequency of e.g. 25kHz and above, as with most 96er systems, there is not much relevant distortion anymore (you could hear) caused by the filter curve. The plus of aliasing which occurs, when going with only 96 instead of possible 192 is less than 20% of the signal for a white noise primary source, meaning you have still 80% of the remaining aliasing problem when increasing from 96 to 192 (and possibly still 50% when increasing from 192 to 384 and so on).

But remember: This aliasing is very small, distributed over the whole audible spectrum and causes "digital noise" in the range of less than 0,1% of a typical audio signal. For example the 96er recording came with 0,04% of AA-noise, 192er system with 0,03% - wow. For rare cases when recording signals with only high frequencies like high drums, it might be 10 times worse. Anyway the improvement is small

From the point of audio recording 192kHz is not necessary. It has some meaning when doing real virtual analog modeling with true oscillators with fine granularity as I investigated here: VAM

One can recalculate that the mathematical way, that higher sample frequencies lead to less error accumulation in real oscillation filters, where it is another question if people can here, that a filter drifts quicker away from reality as another, since nobody can compare it to the reality because there is none available.

There is absolutely some theoretical advantage performing the recording in 192kHz with a ADC having an optimized AA-filter at e.g. 30kHz edge frequency and very low distortion in the listening band and an acceptable ripple in the stop band and then post processing this with a digital filter later. This means:

Maintain the noise in the signal caused by the filter ripple and use it as dithering for the algorithms. Then down sample mathematically at the very end of the production. But again all gear in the signal path must be optimized for this higher bandwidth to make use of it's advantage. Especially the algorithms must be totally clean for this and operate at the higher bandwidth and most best accuracy - but the most software (VST) devices always make a tradeoff between calculation speed and accuracy anyway, because they want to save CPU power, so the discussion often makes no sense at all.

Conclusion and Summary

There is no need to invest into 192kHz rapidly. Current ADCs at 96kHz already have very good 7pole hyperbolic analog AA-filters included. They perform dithering and noise shaping too, in order to overcome some sampling limits. So the errors and non linearity are minimized already. If you can get the option for free, take it. But be aware: Some ADC chips do offer 192kHz by simply cutting down the oversampling: Typically audio ADC'S already perform e.g. 256 times oversampling at 96kHz in order to support AA-filtering effectively. Using the same AA-filter just with 192kHz x 128 is not better, but slightly worse.

If you want to make use of 192kHz: Get a newer, 192kHz only ADC and perform down conversion at the end of the process offline with a software tool using look ahead mechanisms - in terms of FIR filters with a large number of TAPs and strong overlay and optimized windowing. Processing e.g. a 192kHz stereo stream with effective FIR-Filters with e.g. 4096 TAPs (which was not too much for this quality demand) requires a calculation speed of nearly 1,6Mio MUL operations / secs which hardly can be done in a nowadays VST plugin. This has to be done offline with appropriate software or special real-time hardware like FPGAs.

Read about an example FPGA platform for 96kHz audio processing.

Read more in the earlier article about the advantage of 96kHz.