posted 27 April 2003 02:11            

ultimate_042603.kym

What I find is that the real action in terms of signal delay occurs below 1 KHz and in this region nearly all the circumaural headphones behave alike. To my astonishment I found that the phase droop in these, even good ones!, headphones is such that signals at 1 KHz and higher are delayed by as much as 20 ms with respect to 20 Hz tones. That's on the verge of becoming distinct delivery.

So I sat down and constructed an FIR inverse all-pass filter to compensate for this phase droop. When I did that, I got very good phase response, but the whole bottom dropped out. Hence a low-shelf filter is placed ahead of the phase droop preemphasis that very closely cancels the amplitude drop from the all-pass filter.

[All-pass FIR filters will exhibit this non-ideal behavior simply because of finite sample rates. Longer FIR filters perform better than short filters, assuming infinite precision arithmetic. In the end, we make a compromise with 24 bit arithmetic, and reasonably short filters to capture most of the action.]

Plugged into the output L and R channels of Ultimate and voila! One astonishing listening experience. Gone is the old 4-pole inverse of a model -- too much bass and high frequencies. That bass was luscious but ridiculously overpowering. Now we have a clean vista, and the clarity in the delivery of sounds is breathtaking indeed! So much was locked away in those recordings that I never realized was even in there. Now with the headphone phase droop preemphasis filter --- I can only say, give it a try. You will be astonished!

posted 27 April 2003 12:35            

But the Emperor appears to have no clothes... I get so caught up in the excitement of discovery that I must trick myself into believing that I hear remarkable improvements, when in fact, none exist at all...

Sorry,

posted 28 April 2003 02:10            

ultimate_stereo_042703.kym USERPROG.zip

Attached are the newest phase droop preemphasis filters and the Stereo listening environment that takes advantage of them.

This is not "perfect" but very close. First of all I was reading the group delay characteristics of headphones upside down... Bass is actually delayed into the headphones, not advanced. The delay is as much as 20 ms at 20 Hz, dropping to zero around 100 Hz and higher.

This new filter is composed of a 2-pole parametric EQ filter at 20 Hz, cut by 20 dB, and Q = 0.5, cascaded with a low-shelving filter with a boost of 12-20 dB at 80 Hz, Q = 0.5. The shelving filter at 12 dB boost gives flat bass response with a -3 dB point around 40 Hz. At 20 dB boost, it provides a broad peaking bump of around 4.7 dB at 60 Hz, with a -3 dB point down well below 20 Hz.

The filter takes advantage of the group delay characteristics of the EQ cutting filter, which actually advances its center frequencies by a considerable amount. Instead we use the high-side wing of the filter to produce a gradually decreasing group advance from 20 Hz to around 50 Hz. But it alone would severely cut the bass response, so we compensate by reemphasizing the bass with the shelving filter. This in turn cuts the zero group delay point to around 40 Hz. That's okay, because by then, the headphones are only delaying by around 3 ms. So while not quite perfect, as would be had with a very long FIR filter, this is very close to ideal.

The result does produce extended bass response, even with no extra boost above 12 dB from the low shelving filter. Adding a little more boost helps overcome the deficiencies of our hearing as shown by the Fletcher-Munson curves in the low bass region.

[We are all practically deaf below 50 Hz -- it's a good thing too! Because my SPL meter on C-weighting indicates a rip roaring noise level down below 60 Hz, running around 65-70 dBSPL. We would be deafened by all the rumble down there were it not for our poor hearing in the ultra bass region! But for musical pleasure, adding a little ultra-bass boost can be quite nice.]

So, given that bass is delayed out of the headphones, I have to wonder about BBE technology? The delay of the bass corresponds more closely to what I have read elsewhere, and accounts for some of the success enjoyed by the "pulse stretching" fans, where you lowpass filter the bass with a tight filter, say order 4 Butterworth at 50 Hz, and then add to a delayed version of the original signal, with about 10 ms delay.

That "pulse stretching" produces almost the same amplitude characteristics as a low-shelving filter, but the group delay character is quite different. I happen to like my phase droop preemphasis filter more than the sound of the "pulse stretcher", but both are quite nice.

To use this newest version of the stereo listening environment, you need to run my microsounds included in the Zip file. In particular, you need the microsound named DirectIIR. This gives us the ability to make 2-pole filters and control Q, cutoff, and gain as needed for all kinds of filters. My low-shelving filter and the peaking EQ filter are both variants of 2-pole filters. So you could say, that my phase droop compensation filter is a 4-pole solution.

Enjoy!