Feedback heads

[Self-lather]

I'm trying to get a good technical understanding of feedback, what it is, and how it works. I've been scouring the forum and the web for info, but I haven't found a good description of it.

Anyone care to take a crack at explaining it, or point me towards a good resource for understanding it?

Thanks

-Thomas

[markrob]

Hi,

I'll give it a shot.

In an open loop system, you send a command signal to the head from your power amplifier that tells the head how to move the cutting stylus. If the head were a perfect device, it would convert the command signal into groove motions that when played back, are an exact reproduction. Because of physical and electrical limitations, this never occurs. You can correct for this by measuring the difference between what you want to happen and what really occured. You can then adjust the command signal to compensate for the devation in advance of recording. EQ would be one possible compensation you could apply.

In a feedback head, there is a sensing element physically mounted on the head as close as possible to the stylus. This element measures the actual response of the head to the command signal. This response signal is compared with the original command signal in real-time and the difference between them is used to make corrections to the head drive. The result is a system that more closely follows the desired command signal.

In the case of a moving coil head, the command signal is the voltage provided to your power amplifier. The signal represents the desired stylus velocity and is fed to a drive coil on the head. This causes motion of the stylus. Also on the head, is second similar coil that reads the actual motion of the stylus. The coil produces a voltage that is the sensed stylus velocity. The feedback loop is closed by electronic circuitry that calculates the difference between the drive signal and the sensed motion feedback and corrects for any error. By closing the loop, the head is corrected against both static and dynamic errors in the head that might occur due to frequency response errors, resonance(s), temperature, aging, changes in the material being cut, etc.

This is a gross simplification of what goes on, but it should give you the basic idea. As you might guess, the devil is in the details.

Mark

[JayDC]

I think most modern feedback systems use phase cancellation as the primary way of removing the difference..

[markrob]

I think most modern feedback systems use phase cancellation as the primary way of removing the difference..

Hi,

I'm not sure what you mean. I've never seen that term used to describe a feedback control system.

[TotalSonic]

I think most modern feedback systems use phase cancellation as the primary way of removing the difference..

Hi,

I'm not sure what you mean. I've never seen that term used to describe a feedback control system.

Phase cancellation means that a signal whose polarity is reversed (180 degrees out of phase) relative to a source will in fact cancel out the signal of the original completely when they are blended together at the same amplitudes.

Best regards,
Steve Berson

[markrob]

I think most modern feedback systems use phase cancellation as the primary way of removing the difference..

Hi,

I'm not sure what you mean. I've never seen that term used to describe a feedback control system.

Phase cancellation means that a signal whose polarity is reversed (180 degrees out of phase) relative to a source will in fact cancel out the signal of the original completely when they are blended together at the same amplitudes.

Best regards,
Steve Berson

Hi Steve,


I know that, but I don't understand how this applies to a feedback control loop.

[TotalSonic]

I know that, but I don't understand how this applies to a feedback control loop.

From my understanding some amount of the feedback signal with its polarity reversed is mixed into the original source signal and this blend is in fact what gets sent to the cutterhead. That way if the original source is causing a large resonance in the cutterhead this would get somewhat canceled out.

Best regards,
Steve Berson

[JayDC]

I think that it works somewhat like this, off the top of my head, and based on no real research on the topic, lol:

The feedback carrier is blended with the carrier to create a difference, then the difference is subtracted from the carrier, with no latency.

How is this achieve, who knows, I bet flo does..

that is my uneducated theory based on what I know about sound. Not sure how you would achieve this electrically, or if its even a correct and plausible explanation..

[markrob]

Hi,

I see what your are trying to say. That is exactly the process I was describing. When you subtract the feedback from the command you are in effect inverting the polarity of the feedback (A - B = A + -B). This is often incorrectly called phase. On many mixing boards, there is a phase reverse switch. This is not strticly speaking, a phase reversal; its a polarity reversal. If you did not do the polarity inversion of the feedback signal, you would have a positive feedback system (command added to feedback) and it would likely be unstable. Most common control systems use negative feedback, but it is possible to design a system using positive feedback. The math details of all of this, get heavy, so I did not try to get into that.

Mark

[Self-lather]

Hi,

I'll give it a shot.

In an open loop system, you send a command signal to the head from your power amplifier that tells the head how to move the cutting stylus. If the head were a perfect device, it would convert the command signal into groove motions that when played back, are an exact reproduction. Because of physical and electrical limitations, this never occurs. You can correct for this by measuring the difference between what you want to happen and what really occured. You can then adjust the command signal to compensate for the devation in advance of recording. EQ would be one possible compensation you could apply.

In a feedback head, there is a sensing element physically mounted on the head as close as possible to the stylus. This element measures the actual response of the head to the command signal. This response signal is compared with the original command signal in real-time and the difference between them is used to make corrections to the head drive. The result is a system that more closely follows the desired command signal.

In the case of a moving coil head, the command signal is the voltage provided to your power amplifier. The signal represents the desired stylus velocity and is fed to a drive coil on the head. This causes motion of the stylus. Also on the head, is second similar coil that reads the actual motion of the stylus. The coil produces a voltage that is the sensed stylus velocity. The feedback loop is closed by electronic circuitry that calculates the difference between the drive signal and the sensed motion feedback and corrects for any error. By closing the loop, the head is corrected against both static and dynamic errors in the head that might occur due to frequency response errors, resonance(s), temperature, aging, changes in the material being cut, etc.

This is a gross simplification of what goes on, but it should give you the basic idea. As you might guess, the devil is in the details.

Mark

Excellent, this does make sense! I want to devise a sort of DIY feedback system for my homemade cutter head. If I remember correctly, you were doing some similar experiments. Did you ever get that working?

-Thomas

[markrob]

Hi,

I did do some experiments adding feedback to my Presto 1D head. In that attempt, I basically did what the Grampian does. I added a second winding that allowd me to close the loop around the magnetic flux generated in the head. So, the feedback tended to correct only for low frequency effects (mangetic saturation, and the L/R rollof of the head below 500 hz). I started to play with adding true motional feedback to the 1D, but never got very far. Instead, I've been gearing up to re-visit my DIY stereo head.

I've also been doing experiments with my Grampian head. I had been using it open loop and decided to see what feedback would do for it. I've discovered that you need to add some special circuitry to properly achieve the 20db of feedback that is specified by the manufacturer. Flo's simple Jamaika pre is missing some key elements that prevent this from happening. The original RA-7 or Gotham amplifiers had specific compensation built in to provide this. It was sold as a complete system. If you try to close the loop on Grampian head with a simple summing stage as the Jamaika pre does, you will not take full advantage of the head. You'll be lucky to achieve 5db of feedback and stability will be marginal. I plan to post my results soon.

One thing to keep in mind is that you have to have a very good understanding of the open loop gain and phase characteristics of your head before you attempt to close the loop. You want the open loop response to be free of secondary resonances and high frequency poles that cause exess roll-off and/or phase shift. Otherwise, you will have a very hard time closing the loop. The other major problem is finding a way to add sensing that is a true representation of the head's motion. It took some clever engineering by Neumann and others to add a sensing coil that does not pickup up crosstalk from the large magnetic fields present in the head. I've been looking at using the Fairchild RF method to avoid this. Finally, you need a firm grasp of the math and access to some measuring tools to get it right.

Mark

[petermontg]

AES E - Libary have a ton of papers on feedback systems

anybody got a membership???

[Amp Doc]

Well it starts as simple as audio is sent to the head... The feedback coil is directly connected to the drive coil and genarates a signal with all the "errors" of the head... Then the feedback is connected to the input of the power amp 180 degrees out of phase thus "Correcting any errors of the head" BUT

It all seems to be about the way the heads feedback works... the grampian head seems to act more like a transformer then a feedback coil for instance... At some points in the frequency spectrum I seen the feedback signal be 180 out of phase and some places be in phase with the sorce signal... so can cause oscillation and damage the head.... A good way to grasp how feedback works is look up "Desinging High Power Audio Amplifers" there are many books out there... Nearly all power amp use feedback to keep them stable And linear and easy to google as a start....

[markrob]

Well it starts as simple as audio is sent to the head... The feedback coil is directly connected to the drive coil and genarates a signal with all the "errors" of the head... Then the feedback is connected to the input of the power amp 180 degrees out of phase thus "Correcting any errors of the head" BUT

It all seems to be about the way the heads feedback works... the grampian head seems to act more like a transformer then a feedback coil for instance... At some points in the frequency spectrum I seen the feedback signal be 180 out of phase and some places be in phase with the sorce signal... so can cause oscillation and damage the head.... A good way to grasp how feedback works is look up "Desinging High Power Audio Amplifers" there are many books out there... Nearly all power amp use feedback to keep them stable And linear and easy to google as a start....

Hi,

I agree, the Grampian works more like a transformer. The feedback coil is not in motion. When you close the loop around a Grampian, the result is flat velocity response at low frequencies. If you drive the head with a standard audio amp open loop, it will rolloff the lows starting around 350hz. This is due to the inductance and DC resistance of the head (1.65 mh and 3.7 ohms).

The transformer action of the head is linear in gain and phase up to very high frequencies. You will see phase shift from power amp input to feedback coil output caused by the L/R time constant. The reason you have instability because there is no controlled dominant pole in the forward loop at high frequencies if you use a flat response power amplifier to drive the head. So as you try to close the loop, the untiy gain crossover is defined by high frequency poles present in the power amp, pre-amp, and head. It becomes a crap shoot where the crossover is and how much total phase shift is present. In my case, if I raised the feedback amount much beyond 5 db, the system would oscillate at 60Khz.

To solve the problem, I designed and built a simple summing amp with a dominant pole at 2.5 Khz in the forward open loop path. This provides a nice -6db/oct roll-off with 90 degrees of phase lag. If you apply 20db of feedback the unity gain crossover is 25Khz. Now you have an nice stable closed loop system that has plenty of gain and phase margin. If you look at the schematic for the RA7 or Gotham PFB-150-WA, you can see that they did much the same thing. If anybody is interested, I can post the schematic BOM, and PCB layout. I have some suggestions on possible mods to the Jamaika pre to add the domiant pole.

Mark

[emorritt]

Mark; I'd definitely like to see what you've come up with if possible - I built Flo's unit three years ago and have changed a few components here and there because it tended to add too much high end. It 'works' and sounds pretty good, but I agree it could perform better. If I recall, the feedback signal from the head is simply mixed back in through a pot to control the level - no other treatment of the signal. Been a while since I've messed with it. Just been too busy with other projects.

[flozki]

great work.
hope you publish your solution. would be interesting..
jamaica feedbackamp is just what i received in one of those studios which they had quite nice sound whit that solution.so i asked what kind of preamps they used back in the days. i never did simulations and even never used this circuit myself.
so all mods approvements are very interesting.

and yes:
the grampian IS a transformer. so only for the low frequency rolloff.
would be interesting if you can reduce THD on higher frequencies . i dont think that you can achieve that with a grampian in feedback configuration.

you apply 20db of feedback at which frequency? how much headroom before it starts oscillating then?

[petermontg]

Well it starts as simple as audio is sent to the head... The feedback coil is directly connected to the drive coil and genarates a signal with all the "errors" of the head... Then the feedback is connected to the input of the power amp 180 degrees out of phase thus "Correcting any errors of the head" BUT

It all seems to be about the way the heads feedback works... the grampian head seems to act more like a transformer then a feedback coil for instance... At some points in the frequency spectrum I seen the feedback signal be 180 out of phase and some places be in phase with the sorce signal... so can cause oscillation and damage the head.... A good way to grasp how feedback works is look up "Desinging High Power Audio Amplifers" there are many books out there... Nearly all power amp use feedback to keep them stable And linear and easy to google as a start....

Hi,

I agree, the Grampian works more like a transformer. The feedback coil is not in motion. When you close the loop around a Grampian, the result is flat velocity response at low frequencies. If you drive the head with a standard audio amp open loop, it will rolloff the lows starting around 350hz. This is due to the inductance and DC resistance of the head (1.65 mh and 3.7 ohms).

The transformer action of the head is linear in gain and phase up to very high frequencies. You will see phase shift from power amp input to feedback coil output caused by the L/R time constant. The reason you have instability because there is no controlled dominant pole in the forward loop at high frequencies if you use a flat response power amplifier to drive the head. So as you try to close the loop, the untiy gain crossover is defined by high frequency poles present in the power amp, pre-amp, and head. It becomes a crap shoot where the crossover is and how much total phase shift is present. In my case, if I raised the feedback amount much beyond 5 db, the system would oscillate at 60Khz.

To solve the problem, I designed and built a simple summing amp with a dominant pole at 2.5 Khz in the forward open loop path. This provides a nice -6db/oct roll-off with 90 degrees of phase lag. If you apply 20db of feedback the unity gain crossover is 25Khz. Now you have an nice stable closed loop system that has plenty of gain and phase margin. If you look at the schematic for the RA7 or Gotham PFB-150-WA, you can see that they did much the same thing. If anybody is interested, I can post the schematic BOM, and PCB layout. I have some suggestions on possible mods to the Jamaika pre to add the domiant pole.

Mark

Mark

Could you post the schematic Bom, and PCB layout and for the dominant pole?? would be great.

Thinking for awhile to build a feedback system.

regards

[Self-lather]

Mark; I'd definitely like to see what you've come up with if possible - I built Flo's unit three years ago and have changed a few components here and there because it tended to add too much high end. It 'works' and sounds pretty good, but I agree it could perform better. If I recall, the feedback signal from the head is simply mixed back in through a pot to control the level - no other treatment of the signal. Been a while since I've messed with it. Just been too busy with other projects.

Emoritt,

Could you post pics of your cutter head you built?

[markrob]

great work.
hope you publish your solution. would be interesting..
jamaica feedbackamp is just what i received in one of those studios which they had quite nice sound whit that solution.so i asked what kind of preamps they used back in the days. i never did simulations and even never used this circuit myself.
so all mods approvements are very interesting.

and yes:
the grampian IS a transformer. so only for the low frequency rolloff.
would be interesting if you can reduce THD on higher frequencies . i dont think that you can achieve that with a grampian in feedback configuration.

you apply 20db of feedback at which frequency? how much headroom before it starts oscillating then?

Hi Flo,

I applied 20db of feedback measured at 1Khz. The head is flat, open loop in this region. With a 2-2.5 Khz dominant pole, the -20db down point is one decade past the pole location. So that sets the unity gain point at 20-25Khz. The way I set the feedback gain was to monitor the FB coil with a scope while driving the input to my summing amp with a 1Khz squarewave. I made sure that as I adjusted the pot to increase feedback, there was no ringing or overshoot. I targeted a closed loop rise time of 14us. As you add more feedback, you can see the rise time decrease. The bandwidth is calculated by the fomula BW = .35/Risetime. Where Risetime is the time for the squarewave to move from 10% to 90% of full scale (my scope displays Risetime parameter on screen, so its easy to use). I like this method because you can evaluate both high frequency and low frequency cutoffs and keep a eye on stability. Here is a link to a nice explanation of the method.

http://www.kennethkuhn.com/students/ee351/text/square_wave_testing.pdf

As far as reducing the THD at high frequencies, I don't see that happening due to the way the head is designed.

Mark

[Amp Doc]

Just a thought..... I wonder if it would be better not to "sum" the feedback signal with the input of a power amp but to inject the feedback signal into the feedback network of the the output stage??? that would help with the slight phase shift of the input differental amp stage... The thread just had me thinking... Quickly looking at the grampian schematic by naim he seems to apply the feedback to the diff stage with a LC network... Think im right??
To build a mosfet amp with the feedback tied into the diff stage would not be hard and may be a good way of building a "modern" amp for these heads... I will have a look at doing this when I get the time...
I do have a very good, Duel differental mosfet amp schematic I did some time ago what I think would be perfect for this project if anyone needs it. And all the parts are easy to get...