New eq selection request

[Presently42]

I've a request for a new eq selection, though I'm unsure where I should post such a request. In any event, I've been using Fab Filter Pro Q to tune my headphones to various eq curves. I've been using the generic eq from REW, as this allows for the greatest precision. However, it doesn't correctly match Pro Q: I must multiply each q value by root 2, and the bell shapes seem to become cramped as one nears Nyquist using the generic eq.

Might there be a way to include Pro Q's eq algorithm in REW? If not, is there a better method than guessing and trying to make two curves look the same?

 

[John Mulcahy]

Worth seeing if one of the other equaliser selections offers a closer match. Otherwise FabFilter's EQ match might let you do what you want within Pro-Q.

 

[Presently42]

I tried all of the others and found three (I think) eq selections which were closer, but not only still required further calculation, but also only had six or some such number of eq points - a number insufficient to get very close to the curve. As for the eq match, I'd thought of it, but I'm not sure how I'd get audio to play through the desired curve and be heard by Fab Filter without using Fab Filter to make the curve audible

By the way, thanks for writing such a brilliant program!

 

[Presently42]

In addition to this new eq selection request, might I also request that the bass be automatically adjusted? As it is, trying to match my headphone's curve to the chosen (Harman) target results in my having to manually add the necessary bass eq points. I, of course, compensate my reducing the volume in Pro Q by the greatest maximum of the filter curve.

 

[John Mulcahy]

REW does not apply boost below the lowest frequency at which the response drops below the target, to avoid the risk of pushing drivers beyond their excursion limits through excessive LF boost. Safer for any boost that aims to increase LF extension to be applied manually.

 

[Presently42]

Given that the boost has to by applied, that my headphone's curve match the desired target curve, and given that I apply global negative volume which matches the peak level of the added eq filters, why not both suggest the requisite negative volume to be applied (currently a feat achievable only by passing the cursor over the tallest filter peak and finding the maximum value), and also allow for full matching of one curve to another? Sure I can do it manually, but I could also match the curves to each other manually. This programme better than any other I have found, provides a convenient way to do this for me.

 

[John Mulcahy]

I have added a headroom figure on the filters panel, but I haven't changed the restriction on automatic boosting outside the response span. If you want to boost easiest is to apply a manual boost filter before running the target match.

 

[Presently42]

The headroom gain is an excellent addition! However, as the gain is always negative, wouldn't having the figure labelled as such be wise? An unknowing user might accidentally add positive gain, resulting in clipping.

When trying to match my headphones to another set, for example, setting a boost filter before running the matching algorithm won't result in a very close match. I present two examples: the first is the Sennheiser HD800's curve altered to match the Sennheiser HD600 curve. Note that a bass boost or minimisation cannot be applied first, and that the resulting filters make the sub-bass considerably lower than it aught to be. One must unnecessarily add more filters in an attempt to rectify what the algorithm should have already dealt with.

In the second example, the Sennheiser HD800's curve is matched to the Sennheiser HE-1. Whilst a bass boost could be applied, it would result in too much sub-bass and not enough bass - and would crucially lack the dip at around 38 Hz, the frequency at which the skull resonates, and which Sennheiser cleverly minimised in their headphone.

I realise that I am using your programme outside of its intended use case; but it works extremely well. Having the option in the setting to match the bass (with requisite warnings!) would make the programme even more efficient.

 

[John Mulcahy]

I'm not really following those examples. In the first the target is below the measurement, there isn't anything to boost. The second looks like a shelf filter at about 100 Hz would serve to raise the response to the target. The REW EQ algorithm is tuned to deal with response deviations that are amenable to parametric EQ, A different approach would be required for more generalised response matching, something like Banks' Fixed pole parallel filter EQ as used in PORC. That generates responses that require a convolution approach rather than parametric EQ. That was on my list to look at about 10 years ago, but it is a significant chunk of work and PORC did the job.

 

[Presently42]

With the first example, though the target (HD600) is indeed below the measurement (HD800), the filters have caused the bass to go too low by around five dB. In contrast to the rest of the curve, which is matched very well indeed, the bass here is very off. I can only attribute this to the bass not being touched if it exceeds the original quality - including if this is because filters have been applied.

For the second example, is indeed a shelf, as noted - but with some non-trivial deviations which currently require manual alterations: the shelf is non-linear, and has an important dip. Whilst this can be modified after the filter curve is made, far easier would be to have algorithm correct this automatically.

I'll have a look at the PORC. However, I'm hesitant to use non-parametric eq, as para eq enables me to make alterations (especially after around 7 kHz, when frequency response curves become less reliable). Indeed, this brings me to a third feature request - one which I'm not sure can be implemented. As I use REW to match one headphone's fr curve to another's, I'm currently having to manually edit the fr data by hand if I wish to mix more than one fr. For example, I'm currently matching my HD800's curve to the Harman Target curve. However, the Harman Target doesn't account for the necessary dip around 9 kHz, and so proposes an ear splitting peak filter there. I'd love to be able to ask REW to use one curve within a certain frequency range, and another (or more - at one time, I was using three or four different curves, which also necessitated my using two different HD800 curves) curve for a different range.

 

[John Mulcahy]

You could use the trace arithmetic Merge B to A function to create a combined response.

 

[Presently42]

That's an excellent feature! However, as I'd like to take certain ranges of the various curves, it is a bit clunky: it only allows me to take one curve and substitute another from a given point onwards - unless I'm not using it properly!

Furthermore, saving a new curve using the merge B to A results in a .mdat file, which cannot be loaded as the requisite .txt in the house curve, that the difference might be found and a para eq curve made. Or, again, am I just not using it properly.?

One final edit: there is an offset being applied, presumably so that the new curve might be easily seen. However, the new curve then has part of itself offset, resulting in everything above the merge point being offset from the desired curve by the offset amount. Whereas I expect the merge curve to exactly match one curve up to the merge point, and then exactly match the other curve, this does not occur.

 

[John Mulcahy]

The feature is for merging near and far field measurements. It aligns level and phase at the merge point. The result can be exported as a text file, like any other measurement. For a more generalised merge tool without level matching you could probably make something in a spreadsheet to do the job.

 

[Presently42]

Makes sense. I'll stick with just editing text files. It's no biggie.

On the bass issue, I spent at least an hour last night trying to get the bass of the Harman Target curve to match the bass of the HD800 curve. After looking at both the REW wiki and trying several different filters and shelves, I finally realised that the type of shelf I'd manually chosen was wrong. This is such an easy fix: present the option to match the entire range. Add warning, ensure there's sufficient negative preamp, set it as not default - but not putting it, in case a user should misuse it, is removing expected and (if my google search on the issue is any indication) desired functionality.

Finally, and back to this thread's original topic, I asked over on the Fab Filter forum if I might know the equations they use for their eq, that a proper eq selection in REW might be made. I received no reply, other than that the generic REW eq selection is cramped near Nyquist, which Pro Q isn't. Is such an eq selection in REW possible? I went through the current selections, and whilst there does seem to be one or two such, they each have problems which prohibit their usage.

If I'm being picky, it's that's merely because REW has enabled me to enjoy a facet of audiophilia I never thought possible, and which I'm still taking great pleasure in exploring - and encouraging others to do likewise!

 

[John Mulcahy]

It is not an easy fix. As I said above, the REW EQ algorithm is tuned to deal with response deviations that are amenable to parametric EQ. It does not work for generalised response matching.

The StormAudio equaliser has some compensation for filter shape as frequency increases. An equaliser that operates at a higher sample rate will also improve that behaviour - the C-DSP 8x12 setting operates at 192 kHz. Changing pro-Q's gain-Q interaction selection may provide a better match for some equaliser selection.

 

[Presently42]

Ah, the StormAudio one looks promising! Alas, though, it only offers twelve filters, which is simply insufficient. Ah well, looks like I'll have to multiply by root 2 or more for now! Speaking of this, I don't suppose you know a nice equation which gives the amount by which one must multiply between the Cookbook q values and decramped ones? Through trial and error I've gotten sufficiently close - but the perfectionist in me would like to have a nice equation.

Unrelatedly, scrolling in the graph portion of the eq window is impossible with normal scrolling motions and can only be achieved by using the buttons or the scroll bar: it'd be nice to be able to scroll there normally.

 

[John Mulcahy]

Q can be compensated by scaling it by sin(omega)/omega where omega (radians) = 2*Pi*Fc/Fs, Fc = filter centre frequency, Fs = equaliser sample rate.

I don't know what you are referring to as "normal scrolling motions". The graph can be dragged using right click.

 

[Presently42]

Well, that's done the trick! Partially. It seems that the Q value I must put in is the inverse of the equation above, and then multiplied by root 2. Furthermore, this only works if I set my system to use, say, 48 kHz (the upper limit of my poor old USB to DAC interface), and REW to use at least 88.2 kHz. I don't fully understand this, and suspect it has something to do with cramping at Nyquist - but I do know that the values for Q that the modified equation gives are in line with the expected (ie, trial and error found) values. This, then, is what I've been looking for. Now the question: can this be implemented in REW, that I needn't perform further calculations?

As for the normal scrolling motions, I refer to either a mouse wheel, or, in my case, a macbook trackpad. That the graph can be dragged by right clicking is a very useful hidden feature!