tell me your experiences with the elliptical equaliser

[dmills]

Just watch out for the need for an allpass in the M component when doing the second order thing, or it will not sum back together with the right slope, and this actually gets somewhat gnarlier in the case of a shelving design as the phase shift thru the allpass network needs to vary with the shelf setting.
I suppose the might be some kind of second order state variable filter that does the right thing automatically?

My point about dynamic processing, was not that EE made any sense up in the top part of the curve, but that we do frequency dependent dynamics anyway (In the Case of the Ortophon chain, the lowpass corner actually moves as the level rises), so it is possible that the phasing would be less objectionable then we expect. Might be worth writing some matlab to experiment with?

4 Channels, optionally with delay in the control links to match the preview, but also a 2 channel version as plenty of cutting gets done with a digital delay for the preview and in that case there is no reason to not just put the EE before the delay line, this is probably more common then cutting from tape with the preview head.

As to tracing errors, yea, digital delay line with the output clock generated by a DDS or such, or just do it in software post IRIAA, the original stuff tried all sorts of games with switch capacitor arrays and such.

[pizzahotline]

I don't bother with a hardware elliptical EQ. I simply use linear phase Fab Filter Pro Q3 VST in in M/S mode to mono the lower frequencies.

[peter sikking]

you know what, let’s have look at what a traditional elliptical equaliser—
first order filter, 75, 150, 300Hz—does.

first, what does it do to the music? well, turns low frequencies to mono.

how much is it actually reducing the S component?
at 20Hz, the 300Hz EE has taken off 23.5dB (17.5dB for 150Hz EE, 11.5dB for 75Hz EE).

the filter also influences the stereo width above the filter frequency; by 1dB one octave up,
0.5dB 1.5 octaves up, 0.25dB 2 octaves up (e.g. 150Hz EE, 1dB @ 300Hz, 0.5dB @ 424Hz,
0.25dB @ 600Hz).

[peter sikking]

OK, the post above looks confusing, because I was not even half done,
and the graph belongs to the text that follows below—

second what does the EE do for the vertical cutting amplitude?

that is what is in the graph above. the horizontal axis is frequency in
octaves, zero is 50Hz. vertical is decibels.

below 50Hz the cutting characteristic turns constant velocity (because
the RIAA stops compensating) and this shelves off the attenuation
of the EE.

it is super interesting to me to see how much (or, how little)
attenuation the 3 settings provide. btw: 20Hz is -1.33 on the x axis.

another way of looking at the graph is to call it a 50Hz shelf, with
3 attenuation settings. I note 2 useful characteristics:

1. the curves with less attenuation are smoother and less steep.
   less steep equals lower phase shift.
2. the curves with less attenuation don’t reach as much into the midrange.

those traits are (IMO) very much worth conserving in a new design for an EE.

[pizzahotline]

It's fascinating to read your technical info about an EE. I didn't know that it actually boosts frequencies above the cut-off frequency - very cool! Is this considered to be a problem? Also, are you hoping to avoid this in your design?

As I stated in an earlier post I use a m/s plugin to mono the bass. Is there an advantage to using a hardware EE over a software one?

Thanks

[dmills]

Be a little careful here, the 2122us time constant in the IRIAA makes the effect on groove amplitude flatten out as you describe, but it does NOT make the effect on the reproduced audio flatten out because it is exactly compensated by the RIAA curve in the reproducer.
In terms of the reproduced audio the slope continues on down.

I am a little suspicious of your 1dB one octave up, at least in terms of what that does to the separation of a reconstructed LR pair, which is what matters, time for some LtSpice...

Here is the (crude as all hell first order thing I drew)

Ist order EE.PNG

Here is what it does to S when driven with two signals 180 degrees out of phase.

out of phase.PNG

Here is what it does to stereo separation (I am just driving one channel, the other is silent!

One channel driven.PNG

The blue line is the driven channel (Left in this case) which bottoms out at just 6dB down, not too bad, but the green curve is the bleed into the other channel of the stereo pair, which does not hit -20dB until ~400Hz!
It is the fact that the bleed between channels is SO sensitive to the amplitude of the S component that makes these things something that is used only when you must.

Regards, Dan.

[dmills]

No advantage to hardware over software except that some people see all Analogue as an advantage in the marketing game.

[dubcutter89]

Hmm, interesting graphs! But I'm not sure if all the said is really true:

the curves with less attenuation are smoother and less steep.
less steep equals lower phase shift.
the curves with less attenuation don’t reach as much into the midrange.

Your graph shows combined response. One being the EEQ, the other the RIAA, and one more the integral from velocity to excursion...
The last 2 being fixed standards and will be ideally perfectly compensated by the transducer and phono preamp. No phase shift.
The classic EEQ in its basic form is a first order filter having a very well known phase response (for those who are that much into that topic..). An EEQ with 150Hz setting will start phase shift (as a rule of thumb) at 1500Hz, the one with 300Hz setting at 3000Hz. I don't see the "steepness" correlating to "more phase shift". All simple linear behaviour....

I didn't know that it actually boosts frequencies above the cut-off frequency - very cool!

I don't see a boost (?) I also don't think there is one, unless somebody uses a higher order device with some resonance or even goes for additional boosting the mid frequency in the Side Channel to give "more stereo"...

Is there an advantage to using a hardware EE over a software one?

In software everything is cheap, easy, reproducable, with almost all options you can imagine. Also allows to copy paste into your preview, modulation, monitoring, mastering tracks...etc.
Hardware is hardware. If made nice can be nice, and if analog can go into an all analog chain for those of us who are still winding reels on their custom decks

Important - As said I don't want to discourage a new project, just want to give some input for those who read my blabla...

Lukas

[peter sikking]

my messed-up posts from yesterday (damn ‘Submit’ button) and you replies from today made
me realise there is more to it. so I made some graphs.

I made an RIAA curve, that you apply during cutting:

RIAA.png

in this and all the following graphs, the horizontal axis is frequency in octaves (zero is 50Hz),
and vertical is in decibels.

and here is the amplitude characteristic (lateral and vertical) of your cutter head:

cutter.png

apply RIAA to the cutter head:

RIAA+cutter.png

yes, this is just the width/depth of the groove on the record.
because cutter head & phono cartridge and IRIAA & RIAA are complementary,
the record cutting -> playback chain frequency response is fully flat (within practical limits).

now here is a standard elliptical equaliser—first order filter, 75, 150, 300Hz:

EE 75-300.png

yes, what the EE does to the music when played back, including all the nagging
side effects, is controlled by the curves above.

however the reason to use an EE has all to do with the vertical cutting amplitude.
so here is the EE, as applied though RIAA to the vertical cutting amplitude:

RIAA+cutter+EE.png

in grey there is the normal no-EE response curve.

and now I am going to let these graphs sink in…

[markrob]

Hi,

I did some simulations using FLowstone and I get the same results as Peter. Your last graph shows the ability of the EE to reduce the vertical excursions with the EE engaged vs. no EE. I think that's correct.

Here is my simulation of that setup with the EE set for 300Hz. This uses the iRIAA curve and an integrator that will simulate cutter head excursion. (Note: Click on the images for higher resolution, if needed.)

Excursion.jpg

As far as the end how the EE affects the final audio, I did two simulations. I fed the EE with only left channel and looked at the spectrum after electrically summing L+R output of the EE.
You can see from the screen shot below that the response is flat. Note that no iRIAA is used here as the overall cutting system is assumed to be flat (iRIAA -> RIAA)

Electrical Sum.jpg

Under the same conditions, If I add the the spectral magnitudes of the L and R separately, you see a small 1db peak. That's due to not taking the phase into account as the electrical sum would do. I'm not sure if our ears do a vector sum or a magnitude sum

Spectral Sum.jpg

Mark

[dmills]

Nothing you do to the amplitude of S has ANY effect on the electrical sum to mono of the LR pair (makes sense because that is the very definition of M!).

Ears are more complicated then that, being as angle of arrival has a real impact, because the ear is itself an angle dependent filter, and this varies from person to person, much to the disgust of the folks trying to do immersive audio in headphones. Google HRTF for the gory details. Acoustic summing is of course a vector sum.

All of that is however a bit of a red herring when looking for the audible impact of the EE however, what is FAR more interesting is to look at what a 1dB level change in S does to the stereo separation in LR, it is disturbing, even in the context of the poor stereo separation inherent to this medium.

Let L=1, R=0, so something panned 100% left, then M=1, S = 1 and L= (M+S)/2, R = (M-S)/2 gets us back to L,R as original.

Now turn S down by say 1dB, S = ~0.9, so L = 0.95, R = 0.05 which in linear terms does not look horrible, but 0.05 is only -26dB, which is in the same order as the separation in a real vinyl playback chain (This ignores the phase shift due to that 1dB of attenuation, which will make this worse!).
Thus an EE is only negligible to the stereo separation above the frequency at which it has <1dB of S component attenuation, so 1 to 2 octaves above its notional 3dB point, that 150Hz EE is still fucking up the separation at 300Hz!

[peter sikking]

btw, you asked if I was sure about -1dB one octave out.

here is a plot of a first order high pass:

roll-off.png

horizontal: octaves, vertical: decibels.

pretty much spot on—
+1 octave, -1dB
+1.5 octaves, -0.5dB
+2 octaves, -0.25dB
+2.5 octaves, 0.125dB

not that I am disputing that you can hear the narrowing one
octave above the nominal EE frequency.

I take that as given and the goal is to reduce it.

[peter sikking]

let’s have a detailed look at that fully-mono to only-one-channel to
fully-out-of-phase transition.

for clarity’s sake that will be done in two separate transitions:
fully-mono to only-one-channel and
only-one-channel to fully-out-of-phase.

I took dmills’ SPICE model and set the EE frequency to 300Hz.
the left input channel is always driven at full level.
the right input channel get attenuated (and phase inverted) to perform the two transitions.

first setup: fully-mono to only-one-channel
the right channel is always in phase with the left one.
the right channel starts at full level, then gets attenuated in 6dB steps to -60dB, then fully silent (zero input).

left channel output:

L mono–one mag.png

the (green) curve at the top is the fully-mono curve, below 300Hz we see how attenuation of the right
channel pushes the level of the left one rapidly towards -6dB. with 18dB of right attenuation (cyan curve)
it is mostly done.

here is the left channel phase:

L mono–one ph.png

the right channel output:

R mono–one mag.png

the (blue) curve at the top is the fully-mono curve, the (green) slope at the bottom is the zero-input,
only-one-channel curve—all left input shunted into the right channel.
in between those two extremes the curves show how below 300Hz things quickly
gravitate towards -6dB and above 300Hz every curve settles at the actual right channel attenuation
(e.g. the curve for -24dB input (grey) settles at about 30kHz at -24dB)

I can confirm that below 30Hz the left and right plots are virtually identical.

here is the right channel phase:

R mono–one ph.png

the more shelving action, the more phase shift. the zero-input roll-off curve settles at 90°.

second setup: fully-out-of-phase to only-one-channel
the right channel is always out of phase with the left one.
the right channel starts at full level, then gets attenuated in 6dB steps to -60dB, then fully silent (zero input).

left channel output:

L out–one mag.png

the (green) roll-off curve at the bottom is the fully-out-of-phase situation.
below 300Hz we see how attenuation of the right channel pushes the level of the left one
rapidly towards -6dB. with 12dB of right attenuation (red curve) it is mostly done.

here is the left channel phase:

L out–one ph.png

already at 6dB of right attenuation (blue curve) the phase shift becomes a much smaller bell.

the right channel output:

R out–one mag.png

below 100Hz the curves are by and large identical to the left channel ones
and above 300Hz every curve settles again at the actual right channel attenuation.

here is the right channel phase:

R out–one ph.png

all the curves in between fully-out-of-phase and only-one-channel transition
from 180° on the far right of the graph to 0° at the left.

with everything changing so rapidly for small attenuations of the right channel,
it really pays to zoom in on that. and that is may next post.

[peter sikking]

it is déjà vu all over again.

I am going to repeat the 8 graphs of my previous post, but this time
the right channel gets attenuated in 1dB steps, down to -20dB.

first setup: fully-mono towards only-one-channel
the right channel is always in phase with the left one.

left channel output:

L+ mono–one mag.png

the left channel phase:

L+ mono–one ph.png

the effects (narrowing, phase shift) are simply increasing
most when close to perfect mono input.

the right channel output:

R+ mono–one mag.png

again the curves below 30Hz for the left and right channel are visually identical.

the right channel phase:

R+ mono–one ph.png

only here there is an even increment of phase shift for every dB of right attenuation.

second setup: fully-out-of-phase towards only-one-channel
the right channel is always out of phase with the left one.

left channel output:

L+ out–one mag.png

the left channel phase:

L+ out–one ph.png

the effects (widening, phase shift) simply are increasing
most when close to perfect out-of-phase input.

the right channel output:

R+ out–one mag.png

a beautiful ribbon.
again the curves below 30Hz for the left and right channel are visually identical.

the right channel phase:

R+ out–one ph.png

also here more change in the phase shift curves when close to perfect out-of-phase input.

I am going to let this all sink in again.

[Radardoug]

Your terminolgy does not make it clear what you are doing. If you have a mono signal, that is a one channel signal. Do you mean you are using two channels, but one is an out of phase copy of the other channel?
As a general rule in these days of digital and mixing to Protools etc. If your mix is badly out of phase at any frequency, then remix it and get rid of the out of phase. Much better than using a filter.

[dmills]

Unfortunately 'very wide' bass is a thing in certain genres, and that generally means out of phase.

Yea, in an ideal world you get handed material where everything is at worst uncorrelated, but negative correlation does happen, and while it is not an issue at high frequency, it absolutely can be one at low frequency.
The reason these filters are so problematic IMHO hides in the green line on the right channel (only left channel driven) plot, where we can see the stereo separation being dramatically reduced up into the mid band.

It would be nice to think that the mix engineers (who are the ones who can sort this) would when mixing for a release to vinyl have an understanding of the limits of the medium, but it just ain't so.

[Radardoug]

Well if you are the cutting engineer, you need to educate them! I do remember in the old days, the cutting engineer would shake his head and send you away.

[gold]

Well if you are the cutting engineer, you need to educate them! I do remember in the old days, the cutting engineer would shake his head and send you away.

Unfortunately the people who wrote those checks to the cutting engineer are no longer writing checks. If you want people who are alive to write you checks its best not to send them away.

[peter sikking]

the graphs from post #13 have helped me to get on solid ground
what the functionality of an elliptical equaliser, that follows my approach,
should be.

starting out with the graph of the depth of the groove on the record vs frequency;
the iRIAA applied to the cutter head:

RIAA &amp; cutter.png

in this and all the following graphs, the horizontal axis is frequency in octaves (zero is 50Hz),
and vertical is in decibels.

now, let’s start by taking out that sub-50Hz rising (caused by the iRIAA):

50 out.png

this is central to my approach.

and now apply the 20Hz practical bandwidth limit for cutting:

the hump.png

I think this shows the operational area (in Hz and dBs) that the second part of
a new EE should cover (the first part being a solution for that 50Hz rising).

frequency-wise there is a plateau here of increased cutting depth between
20 and 500Hz. those are the -3dB points. if we got a low shelf that can
be set, and trusted, to provide attenuation of X dB at Y Hz (and below),
then the -1dB points of the plateau (40 and 250Hz) shows the frequency
range for that shelf.

how many dBs of attenuation? the plateau is 12dB high so that
seems to be a good maximum level.

since most of the time an EE is not needed at all, I expect that
when it is needed, it starts out with needing just a few dB
of attenuation, not a blunt hammer of 12dB, or an endless roll-off.

applying just a few dB of EE is going to go quite some way to
reigning in the nagging side effects.

[jtransition]

If I was designing a new EE I would want the following;
1:Variable frequencies from 0-700hz
2:Ratio control(How much of the content is moved to mono)
3:No reduction of lf content when in circuit.

My current EE (on my desk) has three buttons On / 150hz / 300hz ,it does not reduce lf content and fixes 90% of phase related cutting issues on my lathe (VMS70).