Purify / Low Distortion Filter for (Rigid Cone) Mid-Woofers

[bmc0]

It occurs to me that a simple resister on the driver or a high output impedance amp might not work, reason being that in terms of distortion, it would be bringing the fundamental down as much as the distortion, so there'd be no net gain. You have to bring a higher frequency peak down relative to a fundamental. Am I getting this right?

Not quite. Simply increasing the source/driving impedance while keeping SPL constant will in fact improve certain nonlinearities; namely those which modulate the voice coil inductance. This is because modulating inductance generates a voltage through the coil, which can be thought of as a voltage source in series with the coil. Since it is the current through the coil which determines the generated force (and consequently, the acoustic output), limiting the conversion of this error voltage to current reduces its impact. Here's a simplified model:

Zout represents the source impedance. It can be seen that Zout=0Ω maximizes the current from the error voltage, while a higher Zout reduces it.

Le(x) and hysteresis in the steel parts near the voice coil both cause inductance modulation and their influence can be reduced by increasing the source impedance. Excursion is likely very low in a horn-loaded midrange, so Le(x) is unlikely to be significant. The hysteresis, however, causes inductance modulation even with zero excursion.

 

[Moonlightshadow]

Very cool. Thanks for the measurements.

 

[Tim Link]

This has gotten me thinking about distortion, which is basically overtones. If we can make the fundamental loud compared to the harmonic overtones, then we can reduce distortion. So, horn loading a driver will tend to bring the low end up at 6 dB/ octave. This can be electronically compensated with EQ, and the benefit is that 2nd order should be 12 dB down in best case, and 3rd order 18 dB down. Also, since the excursion of the driver is reduced at lower frequencies, the intermodulation of upper frequencies is also reduced. Other things to consider might be absorptive material that reduces output above the passband, and listening off axis to the driver so that higher frequencies don't reach you as well because they beam more, so you hear less distortion.

One thing I've been doing on my CD horn loaded compression drivers is adding a 1uF capacitor to act as a constant directivity compensation filter. But there's a problem with this because it actually increases distortion. I'm increasing impedance on the lower frequencies more than the higher frequencies. Better to use DSP to compensate for bringing the low end of the driver's range down. For bringing the high end down, use this passive method. BTW, I tested this and sure enough, with the capacitor the distortion is higher than without it. The cap is essentially doing the opposite of this method.

So here's a question: Why not add an inductor to a driver to create a 1st order low pass, creating a 6 dB / octave downward slope across it's operating band. This will require more voltage at the top of the range, but the impedance is higher so the current demand wouldn't be any different, right? And, this should reduce the driver's distortion. Although I don't know what amps think of putting out a lot of voltage, and I don't know how capacitors react to higher voltage thrown at them. Do they get hot? Explode?

 

[bmc0]

One thing I've been doing on my CD horn loaded compression drivers is adding a 1uF capacitor to act as a constant directivity compensation filter. But there's a problem with this because it actually increases distortion.

In what frequency range does the distortion increase? Low driving impedance can be helpful at lower frequencies near the driver's resonance because the negative feedback afforded by the back EMF helps to linearize certain parameters.

Assuming you have a little sensitivity to spare, you can try an R-L (a.k.a "sine-cap") high pass filter like I demonstrated here.

 

[tmuikku]

So here's a question: Why not add an inductor to a driver to create a 1st order low pass, creating a 6 dB / octave downward slope across it's operating band. This will require more voltage at the top of the range, but the impedance is higher so the current demand wouldn't be any different, right? And, this should reduce the driver's distortion. Although I don't know what amps think of putting out a lot of voltage, and I don't know how capacitors react to higher voltage thrown at them. Do they get hot? Explode?

Yeah, why not, if double impedance (of voice coil inductance) at any frequency, the distortion should be cut in half, compensate with EQ. It takes more voltage from amplifier yes, but typically also power drops towards highs so likely lows still dominate how much voltage your amp needs to be able to swing. I assume this was your line of thought, and that it was for a midrange / woofer, and it seems legit. I'm using big inductor in series on my mids and woofers and it low passes distortion and it might sound better, but since the system is cruising anyway (big three way in an apartment) the difference is not readily audible in use, although clearly visible in measurements and kinda audible on a comparison I did although frequency response matching is not perfect so give or take.

 

[Digital_Thor]

I chose KEF drivers, namely for their idea with the rubber suspension between the coil and the midrange cone, to smooth out the resonance.
They call it a "lossy interface". Can be seen on page 14:
https://assets.kef.com/documents/rseries/rseries2018-white-paper.pdf
As woofer, I use the Satori WO24P, which do not break up before 3kHz.... I cross at 500Hz.
Better drivers = more freedom to filter and design your speaker.
But the notch filter is a good way to kick some life into older drivers, and push the useable range

 

[MAB]

If I understand the application, the RLC network's impedance at the resonance frequency of the inductor and capacitor is going to determine how much damping is applied. The damping is the resistor of the parallel RLC network, inserted in series with the driver.

At the circuit's resonance, the impedance is R.

I have a driver that is ideal to demonstrate this, a Seas W18 version 2. It has a large breakup mode at 5.17kHz. It measures just like the spec sheet, although my data is nearfield.

The resonance has implications, distortion products at various ratios of the fundamental. Here is the driver's raw response and distortion, no filter:

Keep in mind, these distortion peaks are almost 60dB below the signal, very hard to hear. This is a very low distortion driver, so the resonances stand out quite dramatically. One thing for sure, the peak at 5.17kHz is really audible, and should be equalized if you use the driver close to that frequency.

I wired a filter as described above with a 0.15mH inductor, and 6.3uF capacitor, with a 100 Ohm resistor to deal with the peak.

The majority of the peak is gone. Sweeps sound noticeably better without the irritating peak.

And the distortion is reduced.

The frequency response peak with the RLC filter is now flattened, and distortion is now free of the most of the odd harmonics, with reduced 3rd HD remaining:

For reference, the filter's measured response:

As dramatic as the measurements are, I'm not sure I can hear the distortion reduction. I sure can hear the flattening of the peaked frequency response. It's a cool way to control the peak and reduce distortion simultaneously.

I've done this same exercise with the earlier version of the Seas W18 driver, and compared to using DSP to flatten the peak of the driver. This application of the RLC filter does provide lower distortion than DSP, due to the 100 Ohm resistor dominating the impedance at the resonance. The resistor is key.

None of this is specific to good or bad, but it does point out these more exotic driver formulations come with their own tradeoffs.

 

[Digital_Thor]

I've only worked with active systems, so please bear with me
But when you make the passive network to counter the break-up. Can you then make the Q higher, so that the notch is a bit more narrow, targeting only the peak, so that you do not get that 2-3dB reduction at around 3kHz?
Further. In my DSP, I can choose between constant and proportional Q, which should help with having less sideband issued when using larger Q values. Could this make a difference too?

 

[olieb]

This and the other thread are very helpful. I am planning on making my R3 active. Instead of throwing out the crossover completely, now I will try to keep the parts that work in the described way. Thanks.

This has gotten me thinking about distortion, which is basically overtones. If we can make the fundamental loud compared to the harmonic overtones, then we can reduce distortion. So, horn loading a driver will tend to bring the low end up at 6 dB/ octave. This can be electronically compensated with EQ, and the benefit is that 2nd order should be 12 dB down in best case, and 3rd order 18 dB down. Also, since the excursion of the driver is reduced at lower frequencies, the intermodulation of upper frequencies is also reduced. Other things to consider might be absorptive material that reduces output above the passband, and listening off axis to the driver so that higher frequencies don't reach you as well because they beam more, so you hear less distortion.

The way I see it the basic mechanism of all these ideas is to reduce out of band output/sensitivity.
One can do this acoustically by horn loading or a bandpass construction (like the Kef mid) and this will work on all kinds of distortion, if the construction does not introduce distortion of its own. But this might restrict your design options.
Or it can be done on the electrical level with a series coil/filter or a sine-cap (series resistor and parallel coil), but this will work on ”electrical” distortion only. This has little effect i.e. for planar drivers or for high excursion distortion.

So here's a question: Why not add an inductor to a driver to create a 1st order low pass, creating a 6 dB / octave downward slope across its operating band. This will require more voltage at the top of the range, but the impedance is higher so the current demand wouldn't be any different, right? And, this should reduce the driver's distortion. Although I don't know what amps think of putting out a lot of voltage, and I don't know how capacitors react to higher voltage thrown at them. Do they get hot? Explode?

Yes, this way one can take advantage of the decrease of content level with frequency by attenuating passively and then EQing back actively. This can help with hiss, too.
But as the cool measurements from @MAB show, with (very) good drivers the improvements might be close to the inaudible range or even below.

 

[Digital_Thor]

^^
If the R3 is a 2018 version - then see the white paper I linked too. They already got rid of the break-up - mechanically.
I made my KEF R3 midrange coax unit from 2018 active. I have no problems with break-up

If you have noticeable hiss from your drivers in an active DIY design. It can help to lower the gain of the amplifier and raise the volume of the pre-amplifier - if the design of your system allows this - better signal-to-noise ratio.
My amplifiers only have around 21dB of gain, whereas my pre-amp have very high output. This removes a lot of hiss, even with horn systems and other high sensitivity speaker systems.

 

[MAB]

I've only worked with active systems, so please bear with me
But when you make the passive network to counter the break-up. Can you then make the Q higher, so that the notch is a bit more narrow, targeting only the peak, so that you do not get that 2-3dB reduction at around 3kHz?
Further. In my DSP, I can choose between constant and proportional Q, which should help with having less sideband issued when using larger Q values. Could this make a difference too?

With the RLC, you are limited in component values. You can adjust the L, C, and R to get a range of simple notches, but not easily tailored to every situation. Some non-ideal behavior due to the resistance of the inductor too, will limit the range of Q, and make the filter asymmetric.

Ultimately, DSP is going to deal with the FR issues way better. But could still be used on top of the RLC filter to compliment. My DSP desktop system uses this RLC plus DSP crossovers and EQ.

 

[Jukka]

If I understand the application, the RLC network's impedance at the resonance frequency of the inductor and capacitor is going to determine how much damping is applied. The damping is the resistor of the parallel RLC network, inserted in series with the driver.
View attachment 417356
At the circuit's resonance, the impedance is R.

I have a driver that is ideal to demonstrate this, a Seas W18 version 2. It has a large breakup mode at 5.17kHz. It measures just like the spec sheet, although my data is nearfield.
View attachment 417360

The resonance has implications, distortion products at various ratios of the fundamental. Here is the driver's raw response and distortion, no filter:
View attachment 417362
Keep in mind, these distortion peaks are almost 60dB below the signal, very hard to hear. This is a very low distortion driver, so the resonances stand out quite dramatically. One thing for sure, the peak at 5.17kHz is really audible, and should be equalized if you use the driver close to that frequency.

I wired a filter as described above with a 0.15mH inductor, and 6.3uF capacitor, with a 100 Ohm resistor to deal with the peak.
View attachment 417370

The majority of the peak is gone. Sweeps sound noticeably better without the irritating peak.

View attachment 417371

And the distortion is reduced.
View attachment 417375

The frequency response peak with the RLC filter is now flattened, and distortion is now free of the most of the odd harmonics, with reduced 3rd HD remaining:
View attachment 417376

For reference, the filter's measured response:
View attachment 417385

As dramatic as the measurements are, I'm not sure I can hear the distortion reduction. I sure can hear the flattening of the peaked frequency response. It's a cool way to control the peak and reduce distortion simultaneously.

I've done this same exercise with the earlier version of the Seas W18 driver, and compared to using DSP to flatten the peak of the driver. This application of the RLC filter does provide lower distortion than DSP, due to the 100 Ohm resistor dominating the impedance at the resonance. The resistor is key.

None of this is specific to good or bad, but it does point out these more exotic driver formulations come with their own tradeoffs.

There seems to be a caveat, the shape and distance of the HD seems to follow the fundamental quite nicely. That is, when you tune the resonance spike down, you naturally get lower absolute HD. Could you post relative distortion graph if it could state the difference better? Or is this the same measurement as in the other thread? There you didn't have 5k spike. (Sorry I'm with a phone again)

 

[Jukka]

A question on the topic. How about higher impedance drivers? Many drivers have higher impedance versions available, at least in the pro brands, so should we prefer them?

The simplest transducer is an AC circuit that consist of a driving amp and the load (with dynamic driver) is a (voice) coil and some resistance from the coil and small amount from wiring. We have discussed about series resistance and impedance in the circuit for AC signal and increasing the driver impedance increases both. Nowadays we also have high gain amps as a norm, so the lesser voltage sensitivity shouldn't be a problem.

PS. Also the induction varies, quite a lot. How about that?

 

[mhardy6647]

Maybe this is why some people like tube amps. Also, perhaps I should be using a resistor on my midrange horn instead of turning down the volume Digitally to match the woofer and tweeter.

Well... I'd suggest (humbly, of course ) that you'd probably be better off using a fixed L-pad as opposed to a resistor -- so as not to interfere with the crossover.

L pad calculator - attenuation dB damping impedance decibel loudspeaker speaker voltage divider - sengpielaudio Sengpiel Berlin

L pad calculator impedance attenuation damping dB decibel loudspeaker speaker voltage divider - Eberhard Sengpiel sengpielaudio

sengpielaudio.com

Two resistors instead of one.

 

[MAB]

There seems to be a caveat, the shape and distance of the HD seems to follow the fundamental quite nicely. That is, when you tune the resonance spike down, you naturally get lower absolute HD. Could you post relative distortion graph if it could state the difference better? Or is this the same measurement as in the other thread? There you didn't have 5k spike. (Sorry I'm with a phone again)

Here is same data with distortion in percentage. Red is unfiltered nearfield measurement, blue is with the RLC filter. (I tried several different capacitor values, ended up with 6.3uF.) The filter really works to control the distortion, whether you can hear the resulting lower HD, hard to say.

 

[Tim Link]

Well... I'd suggest (humbly, of course ) that you'd probably be better off using a fixed L-pad as opposed to a resistor -- so as not to interfere with the crossover.

L pad calculator - attenuation dB damping impedance decibel loudspeaker speaker voltage divider - sengpielaudio Sengpiel Berlin

L pad calculator impedance attenuation damping dB decibel loudspeaker speaker voltage divider - Eberhard Sengpiel sengpielaudio

sengpielaudio.com

View attachment 417427
Two resistors instead of one.

Assuming I'm using a passive network, yes. L-pad please.