I've been thinking about using a magnetic field sensor (hall effect or magnetoresistive) as a way to control an amplifier's idle current for some time now. They mostly were either not sensitive enough, too expensive, or just not a good match (like the electronic compass chips). Just recently I was (again) browsing places like Digikey, Mouser and Newark to see if anything new has come out, and came across a line of current sensors made by Texas Instruments. The cool thing about them is that they put the current-carrying line right over the hall sensor in the chip, so they have much better sensitivity. The most-sensitive model, the TMSC1108A4, outputs 400mV/amp of current. All while achieving electrical isolation from the mag-field sensor. The current-carrying "sense" input's series resistance is 1.8 milli-ohms! The current-sense line is specified to be able to handle 20 amps continuous. Cost: less than $2.
Inspired by this finding, I put together a class A amplifier circuit, using two of these guys to set the idle currents for the positive and negative halves. Here's the top portion:
This particular implementation produces 2 amps of idle current. With the transistors as shown + the OPA1656 this circuit's THD is fairly low driving an 8 ohm load (1KHz/2.8WRMS = ~.002%). Changing the overall temperature from 25C to 50C changes the idle current by 5% -- 2 amps to 2.1 amps.
The spice model isn't good for performing transient analysis, it imposes a huge processing overhead. So for transient analysis I devised a behavioral model using a 1.8 milliohm resistor and a "BV" voltage generator. A comparison, also done using LTspice, showed a very good match between 25C and 50C. The behavioral model sims much faster.
Since the current-sensing input is isolated from the Hall sensor there are many other ways this could be used to achieve idle-current control in a closed-loop manner. One that immediately comes to mind is a variant on schemes that use an opto-isolator in place of the good ole' Vbe mutiplier, but achieved using a MUCH smaller sense resistor.
The right signal-processing circuitry also would permit good control over class AB output stages. This has been done using more traditional approaches --but perhaps better because the sensing function only adds an extra 1.8 milli-ohms for each sensor.
Inspired by this finding, I put together a class A amplifier circuit, using two of these guys to set the idle currents for the positive and negative halves. Here's the top portion:
The spice model isn't good for performing transient analysis, it imposes a huge processing overhead. So for transient analysis I devised a behavioral model using a 1.8 milliohm resistor and a "BV" voltage generator. A comparison, also done using LTspice, showed a very good match between 25C and 50C. The behavioral model sims much faster.
Since the current-sensing input is isolated from the Hall sensor there are many other ways this could be used to achieve idle-current control in a closed-loop manner. One that immediately comes to mind is a variant on schemes that use an opto-isolator in place of the good ole' Vbe mutiplier, but achieved using a MUCH smaller sense resistor.
The right signal-processing circuitry also would permit good control over class AB output stages. This has been done using more traditional approaches --but perhaps better because the sensing function only adds an extra 1.8 milli-ohms for each sensor.
BTW I replaced the above idle current control scheme with a simple Vbe multiplier. Its idle current varied by more than 300mA over the same temperature range. Increasing the open loop gain of the circuit shown above would further reduce its temperature coefficient so what I showed can be improved quite a bit.
This variation didn't change the circuit's THD to any significant degree.
I'm going to try a slightly different scheme, designed to both increase the loop gain and get rid of the Vcc-2.5V/Vee+2.5V output-swing limitation. More on that later....
This variation didn't change the circuit's THD to any significant degree.
I'm going to try a slightly different scheme, designed to both increase the loop gain and get rid of the Vcc-2.5V/Vee+2.5V output-swing limitation. More on that later....
Here's a study circuit I'm experimenting with. This approach produces a 10mA shift in idle current between 25 and 50C (only Q1 was swept over temperature). It also permits the output to swing much closer to the rails.
Sims don't want to converge unless I provide an initial condition on the capacitor C1 that's close to the final operating point. This was problematic when doing the temperature sweep so I just snuck up on the 50C operating point. Increasing the loop gain by increasing R6 became more and more difficult so I stopped tweaking things once I got the delta-I down to 10mA.
This study was done assuming that the TMCS1108A4 would be used for sensing the idle current. It outputs 400mV/A. It costs the same as the 50mV/A version I simulated above.
It isn't necessary to use an OPA1656 in this circuit, I just grabbed a model I'm familiar with. A TL071 would work just fine for this.
Sims don't want to converge unless I provide an initial condition on the capacitor C1 that's close to the final operating point. This was problematic when doing the temperature sweep so I just snuck up on the 50C operating point. Increasing the loop gain by increasing R6 became more and more difficult so I stopped tweaking things once I got the delta-I down to 10mA.
This study was done assuming that the TMCS1108A4 would be used for sensing the idle current. It outputs 400mV/A. It costs the same as the 50mV/A version I simulated above.
It isn't necessary to use an OPA1656 in this circuit, I just grabbed a model I'm familiar with. A TL071 would work just fine for this.
I wonder if such a sensor and adjusting circuit counteracts the initial signal amplification: if the bias current through the final stage is forced into the controlled value (2 amps in #1), how is it possible to drive more (or less) current into the load?
Should this Hall-current-controller not be limited to the dc level only, and thus adding a pole or zero (depending on the feedback behaviour) to the transfered amplification?
But then, with other strategies like passive rolloff (a polar cap) in the feedback path or an active dc servo, one can be trapped in a very limited low freq bandwidth stabilazation hazard.
Should this Hall-current-controller not be limited to the dc level only, and thus adding a pole or zero (depending on the feedback behaviour) to the transfered amplification?
But then, with other strategies like passive rolloff (a polar cap) in the feedback path or an active dc servo, one can be trapped in a very limited low freq bandwidth stabilazation hazard.
The bypass capacitors around the 600 ohm resistors act as low-pass filters so the control current doesn't vary with the output current.
But I was curious about that very possibility so I reduced the value of the bypass capacitors to 1 nano-farad. The main impact was that the amplitude of the second harmonic increased a little bit. So I think the signal supplied by the OPA1656 dominates the circuit's AC behavior, not the idle-current control circuit.
I also checked the FR of the amplifier with the small capacitors and it was flat from DC to over 100KHz, another validation of my hypothesis.
One good reason for sticking with larger capacitors is that they act to bootstrap the opamp's output. It is limited to +/-18V, but, since the driver transistors are very close to +/-2V the capacitor's bootstrap action restores the drive voltage to the full +/-20V peak.
Thanks for the comments!
But I was curious about that very possibility so I reduced the value of the bypass capacitors to 1 nano-farad. The main impact was that the amplitude of the second harmonic increased a little bit. So I think the signal supplied by the OPA1656 dominates the circuit's AC behavior, not the idle-current control circuit.
I also checked the FR of the amplifier with the small capacitors and it was flat from DC to over 100KHz, another validation of my hypothesis.
One good reason for sticking with larger capacitors is that they act to bootstrap the opamp's output. It is limited to +/-18V, but, since the driver transistors are very close to +/-2V the capacitor's bootstrap action restores the drive voltage to the full +/-20V peak.
Thanks for the comments!
There's another reason you want the filter capacitors around the 600 ohm resistors. It might seem equally reasonable to put filter caps around the I-control opamp but that results in a huge turn-on spike for the idle current. That's with the control schemes I have shown so far, anyway. It also is a bad idea to place two capacitors in the control loop because the phase margin goes to heck. The simulations said so, and I believe them in this case!
Here's yet another variation on the theme:
This uses just one current sensor (this particular simulation is using a faster-to-simulate behavioral model). It uses two optoisolators to source the currents that bias-up the output EF's. This version could move the current-control loop's filter capacitor to the opamp side but I chose not to take that approach due to the bootstrapping advantages. Purists might like the fact that it would remove all capacitors from the signal path, though. I'm not counting the 75pF cap needed to keep the circuit from oscillating. It can be an NPO so wouldn't contribute much, if any, coloration to the sound.
I don't think these opto-isolators' transfer ratio is all that tightly specified so a real-world design would drive each opto-iso separately, using a trimmer on one of them so their effective transfer ratios are equal. Going with two current sensors wouldn't eliminate the need for a trimmer so that's a wash.
As in the example shown in post #3, the idle current changes by just 10mA between 25C and 50C. Attempts to further reduce it cause the simulations to time out, likely due to oscillation problems.
Fun stuff!
Mark
This uses just one current sensor (this particular simulation is using a faster-to-simulate behavioral model). It uses two optoisolators to source the currents that bias-up the output EF's. This version could move the current-control loop's filter capacitor to the opamp side but I chose not to take that approach due to the bootstrapping advantages. Purists might like the fact that it would remove all capacitors from the signal path, though. I'm not counting the 75pF cap needed to keep the circuit from oscillating. It can be an NPO so wouldn't contribute much, if any, coloration to the sound.
I don't think these opto-isolators' transfer ratio is all that tightly specified so a real-world design would drive each opto-iso separately, using a trimmer on one of them so their effective transfer ratios are equal. Going with two current sensors wouldn't eliminate the need for a trimmer so that's a wash.
As in the example shown in post #3, the idle current changes by just 10mA between 25C and 50C. Attempts to further reduce it cause the simulations to time out, likely due to oscillation problems.
Fun stuff!
Mark
Yep, that's the risk (#4).
Would you mind to show the full circuit (#1, #7) so one can have an 'unmirrored' view?
That's the necessary condition to have this nice contraption working!
Further sims have revealed a necessary change in my latest version. 10uf caps between the optoisolator's base and emitter pins. Otherwise the dynamic currents in the output transistors don't match.
A small resistor + diode clamp works well. No need for exotic parts. Note the resistor is chosen to set the idle current at about half a diode drop, and the diode limits the sense voltage for high current output. This regulates the minimum current of the upper OP so only the lower OP "switches". The idle current here is about 30mA and the THD about 0.001% into 4 Ohms. Note that the idle current is not dependent on the OP transistor temperature.
Here you go...
This schematic has the additional 10uf caps on the optoisolators. If they're not there the upper optoisolator's current is modulated by the drive voltage, not a desirable thing! BTW the particular optoisolators are spec'd to have a nominal 100% transfer ratio -- 1:1.
@28WRMS this guy's THD sims out to .0015% @ 1KHz. Its -3dB point is a little more than 1MHz.
There's a small "oopsie" in the above schematic. There are two 2K resistors connected between the output and 600 ohm resistors. They are remnants of some troubleshooting I was doing, before I figured out that the optoisolators' current was being modulated by the drive signal. They are not needed.
Yes. The bias is going to take some time to recover after a burst of audio. Averaging the OP current over time is not a great plan. The bias needs to react faster than the audio signal. LV has some cool ideas for auto bias including his "tandem" circuit similar to the one I posted, and the base current detection bias. Some time ago there was a DIYA post of a commercial NMOS amp in a receiver that used the R+diode supply sense to servo the OP bias, but I don't remember the details. I've had good results switching the fast drivers and not the OPs.
Also note that newer amps combine R10 and R17 for cross coupling. This improves high frequency shoot-through current by pulling OPs off as well as driving them on. Often a cap between the bases of the OPs is added. It also increases the slew during the cross-over where the OP bases unload the driver.
Even class-D amps have bias issue, aka dead time. An important feature of UCD amps is a resistor that sets the bias/dead-time.
V-fets solves some of these issues, but there is no flawless solution. It's not difficult to get better than 0.01% THD and that's good enough for me. Some prefer about 0.1% deliberate 2nd harmonic, IE "single ended class-A" and JLH.
Also note that newer amps combine R10 and R17 for cross coupling. This improves high frequency shoot-through current by pulling OPs off as well as driving them on. Often a cap between the bases of the OPs is added. It also increases the slew during the cross-over where the OP bases unload the driver.
Even class-D amps have bias issue, aka dead time. An important feature of UCD amps is a resistor that sets the bias/dead-time.
V-fets solves some of these issues, but there is no flawless solution. It's not difficult to get better than 0.01% THD and that's good enough for me. Some prefer about 0.1% deliberate 2nd harmonic, IE "single ended class-A" and JLH.
Reducing the maximum time step to 1e-6 second produced a substantial drop in the calculated THD. for the circuit shown in post #11. It's FAR below what I would expect from an actual implementation. I was troubled by all the higher-order harmonics that were showing up in the FFT's, despite all the screwing around I was doing. So it mostly was simulation "noise".
Somehow I sense a contradiction here... not conclusive, yet.
The optocouplers requires a lot of adjustments to serve their purpose - the bias current is measured at the +supply only.
Can you expand on this?
The bias servo will ~clip when driven out of the idle current range. The driven current is just going to drive the bias to zero. If you integrate the current, including driven current, the bias stays at zero for a time after the drive is gone. Simulating a steady state signal does not illustrate the issue.
V-fets have a zero or negative thermal co-efficient, so they are not prone to thermal run-away. However, that may not be completely true for some devices.
V-fets have a zero or negative thermal co-efficient, so they are not prone to thermal run-away. However, that may not be completely true for some devices.
Attempt to reply to post #15:
For class A, you can control the sum of the currents through the two output devices, while the output signal depends on their difference. Simple orthogonal loops.
For class-(A}B, you would have to control some suitable non-linear function of the two currents, usually some smooth approximation to a minimum selector.
For class A, you can control the sum of the currents through the two output devices, while the output signal depends on their difference. Simple orthogonal loops.
For class-(A}B, you would have to control some suitable non-linear function of the two currents, usually some smooth approximation to a minimum selector.
Are you sure?
This is an IRF530 (vertical FET):
It looks like that could easily run away thermally.
This, on the other hand, is a lateral FET (ECW20N20):
Above about 200 mA the temperature coefficient changes and thermal runaway is prevented.
Tom
That's obvious, but that very difference is also inside the bias contol loop, and thus counteracting the signal ''drive'.
I expected a sort of low pass effect on the bias control to valid your statement (aka somesort of servo), but's not present.