Current Control vs Voltage Control: A Large Scale Study

[siarez]

TLDR: This post reports the effect of current driving speaker drivers on a large number of samples. I also show that a driver on current drive performs on par with drivers that are 6x more expensive in terms of odd order HD.

Back Story
~2 years years ago I came across a paper showing the benefits of current drive and then I discovered Esa Meriläinen's book.
I tested the idea on a few drivers and saw significant improvements in odd order harmonic distortions (HD).
This prompted me to seriously consider developing a proper current drive amplifier that I can use in my projects
But before spending too much effort I wanted to make sure that current drive is effective on most drivers, not just a few that I happened to try it on.
So I developed a measurement method (software, hardware, protocol) and tested >30 drivers.
That was a major rabbit hole and took a few months. This is the result of that effort. Hope you find it interesting

The Measurement Method
I don't have a fancy Klippel setup or access to anechoic room, so the measurements had to be very near field.
This means SPL will be high. Most measurement microphones have higher distortion than drivers at high SPL. After going through many mics, I settled on NTI M2010.
All drivers are mounted on a 16x16 baffle. The mic is placed 1 inch from the panel co-centric with the driver.
I measured harmonic distortion using exponential sweeps i.e. Farina method.
All measurements are pre-compensated. Meaning the sweeps are adjusted such that the acoustic output is all flat from 200Hz to 4kHz at 105±0.1dB at the microphone.
105dB at 1 inch roughly translates to ~76dB at 1 meter which I thought is reasonable listening level in the bandwidth I'm testing at.

Side note​

I know this deviates significantly from how speakers are measured typically. Here is the justification: As you'll see in my sample set I have drivers of different sizes, but I still like to be able to compare them "fairly". A small driver needs to work a lot harder to produce the same SPL than a larger driver in the far field, because it is less acoustically efficient. But the small drivers are not expected compete with large drivers at SPL. The goal of my measurement is to measure motor distortion at the (far field) SPL values that are can be expected from each driver given its size. By moving the mic very close, I even out the playing field to a large extent which makes the comparisons across different sizes more fair.

All the measurement where done with QA403 using its Python API to automate the process and I did the necessary signal processing to pre-compensate the sweeps using the Scipy library.
Each driver was measured with the voltage drive (VD) and with the current drive (CD).

The Amplifier
A Texas Instruments TPA3255 amplifier board (in BTL configuration) with post-filter feedback (PFFB) was used to power the drivers. One channel was modified to incorporate current feedback. The current feedback channel achieved an output impedance of 60 ohms and transconductance of 1.75 A/V.
Each driver was measured once with the voltage drive (VD) channel and once with the current drive (CD) channel.


The Drivers
There are > 30 drivers in total.

Brand	Models
Tang Band	W8-1808, W4-1757SB, W4-1720, W3-881SJ
Purifi	PTT4.0X
Fostex	FF105WK, FE126En
Visaton	B100, GF200, B200
SB Acoustics	SB20FRPC30-8, SB20PFCR30-8, SB12NRX, SB12PAC25-4, SB12PFC25-4
Vifa	NE95W-04
Fountek	FR88EX, FR88, FW135F, FW100B
Peerless	TC9FD-18-08, TG9FD10-04, 830656
Seas	MU10RB-SL
Dayton	RS100-8, LW150, RS125S-8
FaitalPro	6FE100
Polk	T15 woofer
Klipsch	Quintet MidWoofer
Wave Core	WF090WA01-01
Multicomp Pro	55-5650, 55-5670

side note: My sample of W8-1808 had a buzzing issue, hence the large peak in distortion.

Results
I have attached all the harmonic distortion (HD) sweeps. As an example you can see the SB Acoustics' SB12PAC25-4 plots below.
Dotted lines are voltage drive and solid lines are current drive. This driver shows very significant improvement on CD.
On average drivers showed an 11dB drop in H3 and 6dB drop in H5.

Analysis
Most drivers see significant improvements in odd order harmonics, but for most people these dB values are meaningless.
So I decided to do a simple and crude analysis to relate the improvements to dollar values. This is the plot below. Let me explain:
First let's ignore the orange dots/lines and focus on blue. Each blue dot represent a driver's measurement on VD.
We have three plots for H2, H3, and H5. The y-axis is the relative (to fundamental) distortion.
The x-axis is the price divided by weight and then normalized so the cheapest driver sits at one. x-axis is in log scale so that distances correspond to price multiples.
Yes, dividing by weight is a very crude way of eliminating size as a confounding factor ...
But it is good enough to show us a clear trend: More expensive drivers tend to have lower distortion and we should all be glad! This trend can be seen by the blue trend line.

Now the interesting bit.
The orange dots are the measurements for those same drivers on current drive.
We see the CD trend line has shifted down significantly. Furthermore the horizontal distance between CD and VD trend lines corresponds to a ~6x price multiple!
This means a driver on current drive performs on par with drivers that are 6x more expensive in terms of odd order HD - on average.
I know this is very crude but I believe this puts an educated dollar value on the improvement you get from current driving speakers.

Conclusion
I think by showing the efficacy of CD on a variety of drivers, this data strengthens the case for current driving speakers significantly.

 

[KSTR]

Absolutely breathtaking! Big Thank You for the all the time and effort that went into this professional survey and sharing hard data with the community.
Nice to see that even a Purifi can benefit from current drive and also nice to see there are some real gems in the fullrange department.

 

[siarez]

Absolutely breathtaking! Big Thank You for the all the time and effort that went into this professional survey and sharing hard data with the community.
Nice to see that even a Purifi can benefit from current drive and also nice to see there are some real gems in the fullrange department.

Thanks for the kind words. Yes there are some gems in there. To me the Multicomp Pro 55-5650 stands out as offering amazing value.

 

[DVDdoug]

All measurements are pre-compensated. Meaning the sweeps are adjusted such that the acoustic output is all flat from 200Hz to 4kHz at 105±0.1dB at the microphone.

That's a problem and it's THE problem with "simple" current drive.

"Pure current drive" fouls-up frequency response badly (depending on the impedance characteristics of the speaker). Speakers are designed and tested with a (low impedance) voltage source. (Some tube amplifier fans like what the higher impedance does to the speaker's frequency response so they might say "enhance" rather than "foul-up".)

If you compensate with EQ, you are EQing the voltage and current so your current is no longer "constant" (independent of impedance) and you're essentially back to voltage drive, if your EQ happens to be perfect.

Ohm's law holds and with a given impedance you can't change current without changing voltage.

Most "current drive" speakers have some kind of motion-tracking feedback or other tricks so the built-in amp isn't a "simple" linear transconductance amplifier .

 

[voodooless]

If you compensate with EQ, you are EQing the voltage and current so your current is no longer "constant" (independent of impedance) and you're essentially back to voltage drive, if your EQ happens to be perfect.

And yet, this research seems to show that there still is a significant difference in predominantly H3 distortion.

 

[siarez]

If you compensate with EQ, you are EQing the voltage and current so your current is no longer "constant" (independent of impedance) and you're essentially back to voltage drive, if your EQ happens to be perfect.

This is actually false. I think you are confusing "constant" current w.r.t. frequency v.s. "promotional"to the input signal. I think this is a common point of confusion.
I'd put it this way: The benefit of current drive does not come from how it affect frequency response. It comes from the fact that the driver "sees" high impedance from its terminals, so current distortion is attenuated. This remains the case regardless of how you EQ.
I should also add that my measurement (both VD and CD and across drivers) ALL are made to match (±0.1dB) the orange line in the plot below. (The orange line is an actual measurement of a driver by microphone)

Pure current drive" fouls-up frequency response badly (depending on the impedance characteristics of the speaker).

This is also addressable. The main problem is the impedance peak around resonance. You can address that by design the amp such that its output impedance drops around resonance, so it behave like VD around resonance.
In fact, I have designed an amp that does that. You can read about it on the DIY Audio forum here. Also if anyone is interested in the amp let me know.

 

[amirm]

Can you show the no-eq measurements of voltage vs current drive as seen by the microphone?

 

[Sokel]

Isn't that similar with a Kef (paper?project?) discussed at another thread?
(or it was Purifi?)

Current drive, active speakers (very old) project was one of the most enjoyable reads I ever had regarding audio.

Really interesting, it comes up from time to time.

 

[sqrl]

More expensive drivers tend to have lower distortion and we should all be glad!

Although we'd prefer it was the opposite.

 

[voodooless]

More expensive drivers tend to have lower distortion and we should all be glad!

Not so fast.. there might be other (better) correlation. What about size? Voice coil size? TS parameters. Would be interesting to explore a few of these.

 

[bmc0]

If you compensate with EQ, you are EQing the voltage and current so your current is no longer "constant" (independent of impedance) and you're essentially back to voltage drive, if your EQ happens to be perfect.

This statement is completely incorrect. EQ cannot modify the output impedance of the amplifier.

 

[siarez]

Can you show the no-eq measurements of voltage vs current drive as seen by the microphone?

Hi Amir!

If you are asking about the no-eq frequency response, then the blue trace in my previous post is the no-eq FR, and orange is with EQ.

I should also clarify that the "EQ" here isn't a bunch of biquads. I get the compensated (EQed) response by convolving the sweep signal with the inverse impulse response of the driver under test - (plus LP and HP filter to roll it off at 200Hz and 4kHz). So the EQ is like a gigantic FIR filter. The result is the smooth orange plot in my previous post which feels like magic!

If you are asking about the no-EQ harmonic distortion, below is the HD from an uncompensated sweep. If you compare it with the one from my first post, you can see CD performs worse at low frequencies here and worse than VD, largely because the driver has high impedance near resonance so it is being push much harder there.

I can share more of these plots, but the story will be the same. i.e. where the driver has higher impedance, it gets pushed harder with CD and naturally we see higher distortion.
An 8-ohm version of the same driver would show more distortion than its 4-ohm version, just because it will play louder when on CD.
Sometimes HD will still be lower than VD. But the point is: If we don't EQ then we can't make useful comparisons between CD and VD. We are measuring a non-linear effect, so we should match levels.

Ultimately, if you have a target frequency response in mind for your active speaker design, you are going to EQ anyway. To achieve your target FR, your EQ settings will be different between CD and VD, but your distortion will be much lower on CD.

 

[KSTR]

If we don't EQ then we can't make useful comparisons between CD and VD. We are measuring a non-linear effect, so we should match levels.

I fully support the idea that all variables must be factored out except the one to be examined, therefore EQ'ing all the drivers to the same target, and in both modes.... which is sort of brutal of course, notably if a driver has severe notches and peaks.

In my comparisons more than a decade ago, comparing VC vs CD (and mixed drive) per driver, I simply EQ'd terminal voltage at the driver under CD to flat, which is the target curve under VD automatically, with a set of simple IIR filters using LSPcad's optimizer. So, the FR was the natural FR as seen with voltage drive.

 

[charlielaub]

Hey, great study. You controlled the proper variables like SPL in a way that one can draw some meaningful conclusions. Much better then other tests like this! Good job.

Some comments:
It is interesting to see that for some drivers there is a reduction in 2nd order harmonic under "current drive". You didn't really give the specifics about how the current drive was implemented (I did read what you wrote). I assume there is some feedback from a sense resistor, and not just a series resistance added. This would probably explain why 2nd order distortion is also reduced. IIRC simple current drive primarily improves 3rd order distortion and primarily in the midband around 200-1k Hz. Can you confirm the details about your amplifier circuit for generating the current drive during the measurements and comment on 2nd HD reduction?

It seems to me that current drive can indeed clean up some drivers nicely, but these are usually ones with poorly performing motors, often found in inexpensive drivers. Implementing current drive costs some money, so a cheap driver + current driver may not be all that less expensive compared to an expensive driver with voltage drive. It's also really only useful in an active loudspeaker, so this is closing the door on traditional passive crossover based designs. In some instances the current drive did not improve the driver at all (e.g. Purifi PTT4.0X). So you can also just throw some more money at a better driver and not worry about how to properly implement current drive.

Current drive CAN in many cases improve HD and your results have proven that. But the situation is similar to using MFB on a subwoofer to reduce distortion - it can work when the driver is performing poorly with voltage drive, but you can also just buy a better driver and not worry about the electronics.

 

[popej]

Interesting research, that goes against common practice. It sure rise doubts. I have two

First is the volume level. You estimate it to 76dB/1m. Usually we see here characteristics at 86dB and 96dB, which seems to be more practical volume. Could it be, that at higher volume mechanical problems will dominate distortion, and advantages of current drive would disappear? By mechanical problems I mean coil/magnet asymmetry, suspension asymmetry, membrane deflection and distortion. Measurements at higher power would be appreciated.

Second is the amplifier distortion. I estimate, that you operate at about 0.1W power, where SINAD could be in range 75dB-85dB (see reviews of Douk AB or WiimAmp). This is comparable to your measurements of speaker distortion, where you get harmonics down to -70dB or -90dB. Your current drive amplifier has additional negative feedback, could it be, that it provide less distortion, which influence outcome? Both amps measurements would be appreciated.

 

[siarez]

Hi @charlielaub, Thanks for reading. Great questions/comments

It is interesting to see that for some drivers there is a reduction in 2nd order harmonic under "current drive". You didn't really give the specifics about how the current drive was implemented (I did read what you wrote). I assume there is some feedback from a sense resistor, and not just a series resistance added. This would probably explain why 2nd order distortion is also reduced. IIRC simple current drive primarily improves 3rd order distortion and primarily in the midband around 200-1k Hz. Can you confirm the details about your amplifier circuit for generating the current drive during the measurements and comment on 2nd HD reduction?

Yes, I'm still not sure how to explain HD2 improvements for some drivers. Maybe due to Le(x) ...? But I'm not the only one observing this; Esa Merilainen had similar results here.

The amp board is a 3E TPA3255 board with PFFB. (@amirm has reviewed it here) I removed PFFB on one channel and added my own current feedback circuit with sense resistors, opamps, etc. I used the other (unmodified) channel for VD measurements.

It seems to me that current drive can indeed clean up some drivers nicely, but these are usually ones with poorly performing motors, often found in inexpensive drivers.

My results here show the opposite, i.e. all sorts of drivers benefit from CD. But I'd say the improvement are more noticeable with the ones that have poor/average performance to begin with. For me, I could hear the difference when I was doing the sweeps for most drivers, but not for the Purifi for example.

Implementing current drive costs some money, so a cheap driver + current driver may not be all that less expensive compared to an expensive driver with voltage drive.

My implementation needs an op-amp and a few resistors and capacitors. That's in the order of cents - or a few dollars if you splurge on components. So I would say if the objective is lowering distortion, then CD has much higher ROI than spending on more expensive drivers.

It's also really only useful in an active loudspeaker, so this is closing the door on traditional passive crossover based designs.

Yes, passive designs can't benefit from CD. One more reason to go active

In some instances the current drive did not improve the driver at all (e.g. Purifi PTT4.0X). So you can also just throw some more money at a better driver and not worry about how to properly implement current drive.

Actually Purifi PTT4.0X shows clear improvements in both HD3 and HD5; especially between 1-2kHz for HD3. Whether those improvements are audible is a good question.

But the situation is similar to using MFB on a subwoofer to reduce distortion - it can work when the driver is performing poorly with voltage drive, but you can also just buy a better driver and not worry about the electronics.

Again, CD is quite cheap to implement. Maybe not easy for a DIY'er, but for a company with competent engineering, it is a piece of cake. And as my data shows, it is a big bang for the buck.
These results really gave me the motivation to design my own DSP amp. IMO for active designs, CD is really a no-brainer regardless of drivers used.

 

[siarez]

Hi @popej, thank you for reading and your questions/comment

You estimate it to 76dB/1m. Usually we see here characteristics at 86dB and 96dB, which seems to be more practical volume.

76dB is a very rough estimate. I could be wildly wrong, because during testing it was quite loud - much louder than I ever listen to music myself.
That estimate is also highly dependent on driver size. I think ~76dB is what you'd expect from a 2~3" driver in an infinite baffle at 1 kHz.

Could it be, that at higher volume mechanical problems will dominate distortion, and advantages of current drive would disappear? By mechanical problems I mean coil/magnet asymmetry, suspension asymmetry, membrane deflection and distortion. Measurements at higher power would be appreciated.

Yes, ultimately you hit the mechanical limits, and they'll dominate the distortion. The question is where that happens. And that's different for each driver.
I agree that higher power tests would be good to have. Unfortunately, my equipment and test setup wasn't allowing me go to very high SPLs. Maybe in the future ...

Second is the amplifier distortion. I estimate, that you operate at about 0.1W power, where SINAD could be in range 75dB-85dB (see reviews of Douk AB or WiimAmp). This is comparable to your measurements of speaker distortion, where you get harmonics down to -70dB or -90dB. Your current drive amplifier has additional negative feedback, could it be, that it provide less distortion, which influence outcome? Both amps measurements would be appreciated.

The amp board is identical to what Amir has reviewed here. IIRC my current feedback slightly improved on HD3, but slightly degraded HD2. I'll post the amp measurements when I get a chance.
I have a relatively tight window around the harmonic impulses. I also did very long sweeps which average out the noise. These together helped me lower the noise floor of the measurement.

 

[TNT]

Conclusion

I wish you had added to the legends in all these measurements which one is CD and which one is VD... (e.g. VC HD3 etc...) it might be obvious but still... thanks for all the work!

//

 

[JSmith]

Current Drive

ESP Project Pages - High Output Impedance aka current drive. Does it make a difference?

sound-au.com

JSmith

 

[KSTR]

I wish you had added to the legends in all these measurements which one is CD and which one is VD... (e.g. VC HD3 etc...) it might be obvious but still... thanks for all the work!

//

Clearly stated in initial post, dotted curves is VC, solid curves is CD.