Hi Ed,
There is also the Kuroda Amp1 of 2019 mentioned here https://www.diyaudio.com/community/...-implemented-on-blameless.356086/post-6246923 with a link that shows the following circuit
Q16,Q17 are the power diodes using power transistors. R5,R6 10 Ohms across the diodes like the E Van Drecht patent, but still not nonswitching IMO.
But notice the shunting large capacitors C1,C7 across the bias transistors -- which IMO will kill the autobias loop at HF. Otherwise it is very close to the autobias I am using -- apart from the CCS's Q11,Q14, where I use fixed resistors to floating/bootstrapped +/-9V supplies (for less dissipation in the CCS's).
It would be interesting to simulate ths amp without C1,C7 AND with R5,R6 node broken to the output, to make it NSB. Maybe with R5,R6 link broken the autobias spreaders won't need the shunting capacitors? Anyone interested in simulating these mods?
There is also the Kuroda Amp1 of 2019 mentioned here https://www.diyaudio.com/community/...-implemented-on-blameless.356086/post-6246923 with a link that shows the following circuit
Q16,Q17 are the power diodes using power transistors. R5,R6 10 Ohms across the diodes like the E Van Drecht patent, but still not nonswitching IMO.
But notice the shunting large capacitors C1,C7 across the bias transistors -- which IMO will kill the autobias loop at HF. Otherwise it is very close to the autobias I am using -- apart from the CCS's Q11,Q14, where I use fixed resistors to floating/bootstrapped +/-9V supplies (for less dissipation in the CCS's).
It would be interesting to simulate ths amp without C1,C7 AND with R5,R6 node broken to the output, to make it NSB. Maybe with R5,R6 link broken the autobias spreaders won't need the shunting capacitors? Anyone interested in simulating these mods?
Hi Ed,
Shinichi Kamijo wrote "Evolve TLB – Power amplifier with Trans-Linear Bias" https://x.com/nikstankovic_/status/1798217258885410963
His "simple TL bias" circuits look similar:
Fig 25 using bipolar's and Fig 27 using lateral's, Fig 28 test results. In Fig 27 power diodes D1,D2 are body diodes of 2SK3844 MOSFET's. Presumably the bias spreaders are thermally linked to these diodes.
A translated version of the memo is attached.
Cheers, IanH
Shinichi Kamijo wrote "Evolve TLB – Power amplifier with Trans-Linear Bias" https://x.com/nikstankovic_/status/1798217258885410963
His "simple TL bias" circuits look similar:
Fig 25 using bipolar's and Fig 27 using lateral's, Fig 28 test results. In Fig 27 power diodes D1,D2 are body diodes of 2SK3844 MOSFET's. Presumably the bias spreaders are thermally linked to these diodes.
A translated version of the memo is attached.
Cheers, IanH
Here's my first analyser tests, courtesy of our friend The Saint, using the amp in Post 220 and Post 235 (MUR1615 diodes) and biased at 250mA.
At slightly over 1W gives mainly 3rd harmonic at -60dB or 0.1% slightly higher than my earlier sims. No attempt was made to trim the bias for lower 3rd. Notice the higher order harmonics not worth mentioning. Since the 3rd is inaudible at 0.1% this distortion is inaudible at around 1W. BTW the bump at 50Hz is mains hum getting in somewhere (it's not the rails since rectified 50Hz starts at 100Hz) - with a speaker attached I couldn't hear any 50Hz even with my ear next to the cone - so it's not a problem in reality.
At around 12W the 3rd is still 0.1%. Higher harmonics, 5th, 7th, 9th etc fall away at a fast rate (-80dB/decade) which means these harmonics are inaudible if the 3rd is inaudible, which it is, so all harmonics are inaudible at 12W as well as 1W.
This is a great result -- after lots of wondering whether the simulations were to be believed. The frequency response below shows HF roll-off is around 100kHz!
This was with the input capacitor shorted. Notice the output capacitors (2x3.3mF) shows NO LF roll-off here. BTW this is with no AC feedback from the load, only via the rails for DC centering. (The 47k feedback resistor R6 was removed earlier because dividers R3, R19 were 47k's as mentioned in Post 228). Of course the output caps do restrict the maximum output swing (due to clipping) at the LF end. They serve a useful purpose of speaker DC blocking from faults, so a protection board is not needed, keeping the amp safe and simple. Also a single-rail power supply can be used. Next step is to get revised PCB's made using MUR1615's, or parallel EX3BB 3A diodes. It will include an optional input transformer (see post 235) so standard (non-floating) power supplies can be used!
A 2 channel listening test with music using the 2 channel version in Post 235 was as good as my previous listening test way back Post 157 using Schottky diodes. This test was driven to almost clip (35Vpk) and sounded loud enough with my (inefficient) book-shelf speakers. So it sounded clean and enjoyable. And that's with almost no feedback! All's good so far.
At slightly over 1W gives mainly 3rd harmonic at -60dB or 0.1% slightly higher than my earlier sims. No attempt was made to trim the bias for lower 3rd. Notice the higher order harmonics not worth mentioning. Since the 3rd is inaudible at 0.1% this distortion is inaudible at around 1W. BTW the bump at 50Hz is mains hum getting in somewhere (it's not the rails since rectified 50Hz starts at 100Hz) - with a speaker attached I couldn't hear any 50Hz even with my ear next to the cone - so it's not a problem in reality.
At around 12W the 3rd is still 0.1%. Higher harmonics, 5th, 7th, 9th etc fall away at a fast rate (-80dB/decade) which means these harmonics are inaudible if the 3rd is inaudible, which it is, so all harmonics are inaudible at 12W as well as 1W.
This is a great result -- after lots of wondering whether the simulations were to be believed. The frequency response below shows HF roll-off is around 100kHz!
This was with the input capacitor shorted. Notice the output capacitors (2x3.3mF) shows NO LF roll-off here. BTW this is with no AC feedback from the load, only via the rails for DC centering. (The 47k feedback resistor R6 was removed earlier because dividers R3, R19 were 47k's as mentioned in Post 228). Of course the output caps do restrict the maximum output swing (due to clipping) at the LF end. They serve a useful purpose of speaker DC blocking from faults, so a protection board is not needed, keeping the amp safe and simple. Also a single-rail power supply can be used. Next step is to get revised PCB's made using MUR1615's, or parallel EX3BB 3A diodes. It will include an optional input transformer (see post 235) so standard (non-floating) power supplies can be used!
A 2 channel listening test with music using the 2 channel version in Post 235 was as good as my previous listening test way back Post 157 using Schottky diodes. This test was driven to almost clip (35Vpk) and sounded loud enough with my (inefficient) book-shelf speakers. So it sounded clean and enjoyable. And that's with almost no feedback! All's good so far.
Last edited:
Here's another tutorial piece about the autobias circuit I am using. It has been looked at by Shinichi-Kamijo as mentioned in Post 243 (repeated below)
It is effectively a complementary Wilson like current mirror. The RHS Wilson mirror is from Art of Electronics Fig 3.26 circuit D. The X marks the break between the emitters of Q3a and Q3b to allow modulation of the mirrors current with an audio signal. Notice also the Wilson mirror does not need emitter resistors for the power transistors. This is useful for the autobias circuit I am using. It allows the gain (Gm) of the power stage to be higher than conventional Class-AB output stages. Also, the idle current for this circuit is not much affected by temperature changes in the power transistors.
The power dissipated in the power transistors is Vcc times Ic and the power dissipated in the current sense diodes D1,D2 is around 0.7V times Ic which is smaller by a factor of about 100 times, so the diode temperature rise is much smaller and therefore the idle current is less affected by power output changes.
In practice the power diodes temperature rise will be about 1/10th the power transistors temperature change because these diodes need to be on a separate heatsink which is for economic reasons is likely to be about 1/10th the rating of the main heatsink. But even a net thermal change of 1/10th is a big help in Class-AB (for the reason explained next).
Lets look at what happens when we reduce the emitter resistance of a conventional Class-AB output stage, starting with 0.22 Ohms emitter resistors (the smallest typically used for 15A transistors), then 0.1 Ohms, then to 0.05 Ohms (or 50m Ohms). Wingspread plots are shown below where a particular "optimum" idle current is chosen so there is a flat spot of very low distortion "sweet spot" for smallish input voltages:
The LHS plots are with input voltage and the RHS plots are the same but with output current on the X-axis. The RHS plots Gm increases as the emitter resistance is reduced and there seems to be more gain change as the emitter resistance is reduced - it appears we don't get lower distortion by reducing the emitter resistors. But when the higher Gm (with lower Re) is used for feedback we get the same gain variation independent of Re change - so this is not an issue in practice.
But when we look at the RHS we can see the flat spot of very low distortion "sweet spot" for smallish input voltages widens as the emitter resistance is reduced. For the 0.22 Ohms case the flat spot covers +-60mA peak which gives up to 50mW average with very low distortion (effectively a square-law class-A region). With 0.1 Ohms there is +-240mA peak which gives up to 230mW average with very low distortion. With 0.05 Ohms there is +-600mA peak which gives up to 1.44W average with very low distortion. My last tests were done with 40m Ohms added to the power diodes giving quite low distortion readings at around 1 Watt - IMO "low" for an amplifier with only around 6dB of feedback! The idle current was very stable over time with music even when run to an ocasional clip. Without this autobias loop (as in conventional Vbe biasing) this flat spot cannot be maintained with music power variations and that can explain why it is not exploited. BTW The file for generating these wingspread plots is attached.
Lastly, I should mention the addition of the resistor between the emitters of the power transistors (effectively across the power diodes D1,D2) which keeps the power transistors slightly on even when one diode turns off at clip making it nonswitching Class-AB (aka nsb). The keep on current is R/0.7 where R is the shunt resistance - I use 68 Ohms for about 10mA. This addition does not appear to have been used in other similar autobias circuits so it is an important improvement for autobias - it seen first, seen here on diyAudio😉.
It is effectively a complementary Wilson like current mirror. The RHS Wilson mirror is from Art of Electronics Fig 3.26 circuit D. The X marks the break between the emitters of Q3a and Q3b to allow modulation of the mirrors current with an audio signal. Notice also the Wilson mirror does not need emitter resistors for the power transistors. This is useful for the autobias circuit I am using. It allows the gain (Gm) of the power stage to be higher than conventional Class-AB output stages. Also, the idle current for this circuit is not much affected by temperature changes in the power transistors.
The power dissipated in the power transistors is Vcc times Ic and the power dissipated in the current sense diodes D1,D2 is around 0.7V times Ic which is smaller by a factor of about 100 times, so the diode temperature rise is much smaller and therefore the idle current is less affected by power output changes.
In practice the power diodes temperature rise will be about 1/10th the power transistors temperature change because these diodes need to be on a separate heatsink which is for economic reasons is likely to be about 1/10th the rating of the main heatsink. But even a net thermal change of 1/10th is a big help in Class-AB (for the reason explained next).
Lets look at what happens when we reduce the emitter resistance of a conventional Class-AB output stage, starting with 0.22 Ohms emitter resistors (the smallest typically used for 15A transistors), then 0.1 Ohms, then to 0.05 Ohms (or 50m Ohms). Wingspread plots are shown below where a particular "optimum" idle current is chosen so there is a flat spot of very low distortion "sweet spot" for smallish input voltages:
The LHS plots are with input voltage and the RHS plots are the same but with output current on the X-axis. The RHS plots Gm increases as the emitter resistance is reduced and there seems to be more gain change as the emitter resistance is reduced - it appears we don't get lower distortion by reducing the emitter resistors. But when the higher Gm (with lower Re) is used for feedback we get the same gain variation independent of Re change - so this is not an issue in practice.
But when we look at the RHS we can see the flat spot of very low distortion "sweet spot" for smallish input voltages widens as the emitter resistance is reduced. For the 0.22 Ohms case the flat spot covers +-60mA peak which gives up to 50mW average with very low distortion (effectively a square-law class-A region). With 0.1 Ohms there is +-240mA peak which gives up to 230mW average with very low distortion. With 0.05 Ohms there is +-600mA peak which gives up to 1.44W average with very low distortion. My last tests were done with 40m Ohms added to the power diodes giving quite low distortion readings at around 1 Watt - IMO "low" for an amplifier with only around 6dB of feedback! The idle current was very stable over time with music even when run to an ocasional clip. Without this autobias loop (as in conventional Vbe biasing) this flat spot cannot be maintained with music power variations and that can explain why it is not exploited. BTW The file for generating these wingspread plots is attached.
Lastly, I should mention the addition of the resistor between the emitters of the power transistors (effectively across the power diodes D1,D2) which keeps the power transistors slightly on even when one diode turns off at clip making it nonswitching Class-AB (aka nsb). The keep on current is R/0.7 where R is the shunt resistance - I use 68 Ohms for about 10mA. This addition does not appear to have been used in other similar autobias circuits so it is an important improvement for autobias - it seen first, seen here on diyAudio😉.
Ian - Thanks for an interesting post!
I have briefly looked at bias regulators that use resistor-only sensing. The bias regulator reduces the sensitivity to the output transistors' Is and temperature by the amount of feedback applied. However, it also makes the exponential much stiffer, which means that the emitter resistors should be made smaller. That results in lower distortion but brings back the sensitivity to Is and temperature.
It is not clear to me whether building a bias regulator is the best use of feedback. The same amount of feedback could be applied to reducing distortion directly.
I expect that a conventional EF3 with a Vbe multiplier will experience bias current swings of 2-4x depending on how tight is the thermal tracking. This results in a few tenths of a percent distortion, but that is easily reduced to below audibility by applying voltage feedback. The amplifier must have enough thermal margin (easy).
I think that you need to compare the bias regulators against applying the same amount of feedback to voltage gain. Show that regulating bias is a more effective use of feedback.
Ed
I have briefly looked at bias regulators that use resistor-only sensing. The bias regulator reduces the sensitivity to the output transistors' Is and temperature by the amount of feedback applied. However, it also makes the exponential much stiffer, which means that the emitter resistors should be made smaller. That results in lower distortion but brings back the sensitivity to Is and temperature.
It is not clear to me whether building a bias regulator is the best use of feedback. The same amount of feedback could be applied to reducing distortion directly.
I expect that a conventional EF3 with a Vbe multiplier will experience bias current swings of 2-4x depending on how tight is the thermal tracking. This results in a few tenths of a percent distortion, but that is easily reduced to below audibility by applying voltage feedback. The amplifier must have enough thermal margin (easy).
I think that you need to compare the bias regulators against applying the same amount of feedback to voltage gain. Show that regulating bias is a more effective use of feedback.
Ed
Last edited:
Hi Ed,
The short answer is my autobias loop does not subtract feedback from the forward path -- as Malcolm Hawksford said in his autobias/nsb paper an autobias loop can be "made orthogonal to the main amplification process" on page 2 (paper available here https://www.researchgate.net/public...d_non-switching_power_amplifier_output_stages).
Orthogonal is a maths term that means two things are independent of each other.
As far as I have seen the wingspread plots and resulting crossover distortion with my autobias loop is the same as the standard bipolar Class-AB push-pull stage (without EF drivers) -- the difference with my autobias loop being the emitter degeneration resistors can be eliminated (leaving only small parasitic resistances in the power diodes) while still being thermally stable without these resistors.
---
In my recent autobias circuits I use small emitter degeneration resistors (like 50m Ohms) to provide effectively error correction to get a limited squarelaw region low distortion region up to about 1 Watt which is sufficient to cover typical music listening -- and it is stable with temperature variations with typical music. With this "trick" I can use most of the power stages transconductance Gm for voltage gain, giving a very simple power amplifier with acceptably low distortion for typical music use. With a 2 volt rms signal (eg soundcards) most of the gain is used for voltage amplification. I find that very little feedback works well, but others might prefer more negative feedback which can be done in many ways and I'll leave that to others.
---
A quote from Malcolm Hawksford's paper:
A good question. Thanks.
The short answer is my autobias loop does not subtract feedback from the forward path -- as Malcolm Hawksford said in his autobias/nsb paper an autobias loop can be "made orthogonal to the main amplification process" on page 2 (paper available here https://www.researchgate.net/public...d_non-switching_power_amplifier_output_stages).
Orthogonal is a maths term that means two things are independent of each other.
As far as I have seen the wingspread plots and resulting crossover distortion with my autobias loop is the same as the standard bipolar Class-AB push-pull stage (without EF drivers) -- the difference with my autobias loop being the emitter degeneration resistors can be eliminated (leaving only small parasitic resistances in the power diodes) while still being thermally stable without these resistors.
Ed, sorry I can't provide a rigorous proof. So far I only have anecdotal evidence.
---
In my recent autobias circuits I use small emitter degeneration resistors (like 50m Ohms) to provide effectively error correction to get a limited squarelaw region low distortion region up to about 1 Watt which is sufficient to cover typical music listening -- and it is stable with temperature variations with typical music. With this "trick" I can use most of the power stages transconductance Gm for voltage gain, giving a very simple power amplifier with acceptably low distortion for typical music use. With a 2 volt rms signal (eg soundcards) most of the gain is used for voltage amplification. I find that very little feedback works well, but others might prefer more negative feedback which can be done in many ways and I'll leave that to others.
---
A quote from Malcolm Hawksford's paper:
BTW I notice his Figure 3.2 is similar to circuits used above -- except he senses collector current with two 1 Ohm resistors which is impractical IMO. Power diodes in the emitters is simpler and wastes less power for current sensing than 1 Ohm resistors. These diodes do show some temperature changes with output power changes, but this is not an issue in practice when music spends most of its time around 1W despite the ocassional clip.
The bias regulator adds phase shift. The voltage feedback loop cannot have as much gain because some of its phase margin has been used by the bias regulator.
I am not asking for a proof. I want a practical circuit in which a voltage feedback loop that contains the bias regulator achieves lower distortion than a voltage feedback loop with fixed bias.
Ed
I am not asking for a proof. I want a practical circuit in which a voltage feedback loop that contains the bias regulator achieves lower distortion than a voltage feedback loop with fixed bias.
Ed
Hi Ed,
As it stands my nonswitching autobias amps (from Post 202 and on) are no different raw THD than the conventional Class-AB output stages. The bias feedback loop in my nonswitching autobias amps are orthogonal to the forward transfer function -- so the wingspread plots which dictate distortion levels is no different the conventional Class-AB output stages. That is why I posted simulated wingspread plots generated by the analytic equation solutions for conventional Class-AB in Common Emitter (Post 245). An Oops: I'll repost the updated wingspread simulation file to stop the error message when run under LTXVII and later.
You are asking for "a practical circuit in which a voltage feedback loop that contains the bias regulator achieves lower distortion than a voltage feedback loop with fixed bias." So my nonswitching autobias amps do not achieve your aim.
BTW have you considered the Barry Blesser element used in Class-AB? EG from Post 18 and on here https://www.diyaudio.com/community/threads/common-base-input-for-power-amplifier.387450/post-7093872.
But I prefer the simplicity and flexibility and robustness of my nonswitching autobias amps:
Flexibility: They can accept power BJT's or MOSFET's without changing the circuit, just re-tim the idle current. With power MOSFET's you parallel complete amps (PCB's), called 'slicing' in the LT1166 app notes, rather than use the conventional method with emitter resistors for current sharing.
Robustness: Slicing is a more robust method and wastes less power in balancing resistors -- which don't work very well with MOSFET's, even with balancing resistors, manual selection is usually required. SOA current limiting is achieved with the choice of the pull-up resistors to +-9V for the bias loop.
BTW I use 220 ohms with the transistors I have which have Beta's of around 120, but this resistor is changed if the gains are higher or lower to get 4.5A peak output current (but it increases to 5A when the transistors are running hot). The amp can then survive a dead output short at full drive for a few (tens of?) seconds without damage, a polyswitch or fuse can be added to fully protect the amp against sustained output shorts.
As it stands my nonswitching autobias amps (from Post 202 and on) are no different raw THD than the conventional Class-AB output stages. The bias feedback loop in my nonswitching autobias amps are orthogonal to the forward transfer function -- so the wingspread plots which dictate distortion levels is no different the conventional Class-AB output stages. That is why I posted simulated wingspread plots generated by the analytic equation solutions for conventional Class-AB in Common Emitter (Post 245). An Oops: I'll repost the updated wingspread simulation file to stop the error message when run under LTXVII and later.
You are asking for "a practical circuit in which a voltage feedback loop that contains the bias regulator achieves lower distortion than a voltage feedback loop with fixed bias." So my nonswitching autobias amps do not achieve your aim.
BTW have you considered the Barry Blesser element used in Class-AB? EG from Post 18 and on here https://www.diyaudio.com/community/threads/common-base-input-for-power-amplifier.387450/post-7093872.
But I prefer the simplicity and flexibility and robustness of my nonswitching autobias amps:
Flexibility: They can accept power BJT's or MOSFET's without changing the circuit, just re-tim the idle current. With power MOSFET's you parallel complete amps (PCB's), called 'slicing' in the LT1166 app notes, rather than use the conventional method with emitter resistors for current sharing.
Robustness: Slicing is a more robust method and wastes less power in balancing resistors -- which don't work very well with MOSFET's, even with balancing resistors, manual selection is usually required. SOA current limiting is achieved with the choice of the pull-up resistors to +-9V for the bias loop.
BTW I use 220 ohms with the transistors I have which have Beta's of around 120, but this resistor is changed if the gains are higher or lower to get 4.5A peak output current (but it increases to 5A when the transistors are running hot). The amp can then survive a dead output short at full drive for a few (tens of?) seconds without damage, a polyswitch or fuse can be added to fully protect the amp against sustained output shorts.
Last edited:
Thanks for the likes 👍
Here's how I got my +-9V supplies. I added 4 secondaries to the main power supply transformer. I used 0.32mm wire on a 300VA transformer. Each secondary gives 6V AC. Then four full-wave voltage doublers using 1000uF/16V caps, have 7805 regulators with divider resistors giving four isolated regulated outputs at 9V for up to 300mA each. The matrix board is seen in the back left. These are combined into two +-9V supplies for a two channel amp (or a one channel bridge).
This was my first option, which is accessible for DIY without needing extra transformers (or DCDC converters).
My second option is using 3W Vigortronix universal ACDC converters. These can run of mains 240V or 110V AC, or can use DC from the amps main power rail since they work down to a fairly low DC input voltage. They are well regulated and output short protected and cost about the same as many of the DCDC brick converters of the same power. This option is obviously much easier than adding windings.
I measured 9.015V no load with 38V DC input and 9.000V with 0.3A load. The spread in output voltage between units was better than 1%.
Here's how I got my +-9V supplies. I added 4 secondaries to the main power supply transformer. I used 0.32mm wire on a 300VA transformer. Each secondary gives 6V AC. Then four full-wave voltage doublers using 1000uF/16V caps, have 7805 regulators with divider resistors giving four isolated regulated outputs at 9V for up to 300mA each. The matrix board is seen in the back left. These are combined into two +-9V supplies for a two channel amp (or a one channel bridge).
This was my first option, which is accessible for DIY without needing extra transformers (or DCDC converters).
My second option is using 3W Vigortronix universal ACDC converters. These can run of mains 240V or 110V AC, or can use DC from the amps main power rail since they work down to a fairly low DC input voltage. They are well regulated and output short protected and cost about the same as many of the DCDC brick converters of the same power. This option is obviously much easier than adding windings.
I measured 9.015V no load with 38V DC input and 9.000V with 0.3A load. The spread in output voltage between units was better than 1%.
In Post 235 I reported running 2 channels from a single floating power supply.
Here's a simulation showing this with two input signals, 1kHz and the other channel 2kHz. In this simulation generating an FFT of each output shows no cross-modulation, showing the two amps give independent output signals from a single floating power supply.
So it is possible to run 2 or more channels off a single floating supply using this method - which requires all inputs other than the first one to be isolated and all loads apart from the first to be isolated. In Post 235 I used a small 1:1 transformer for the isolation in my demo.
In this sim (attached) the LHS input is grounded via R3 and link. The LHS load 'Out1' is also grounded.
The RHS input (-) is floating since the link for R12 is open and the RHS load is fully floating.
The power rails are floating using a single supply.
The LHS amp uses the divider R1 and R2 for DC centering. The RHS amp uses the divider R4 and R7 for it's DC centering. Therefore when the supply V3 is shifted relative to common by the LHS amp, then the RHS amps DC reference voltage 'b2' maintains its DC reference point.
So the RHS amp is not affected by the LHS amps output voltage changes.
Notice the top table of harmonics does not contain much 2nd harmonic; you might expect to get some from the other channel running at 2kHz.
But both tables have exactly the same 2nd harmonic levels (which probably comes from numerical noise).
I hope this helps to understand how/why the method I got to work in Post 235 actually works.
Further, it allows bridging of 2 channels, as I showed on the bench.
To bridge in this simulation just use "-Vpk" for the input signal to the RHS channel and make both signals the same frequency. Then connect a single load from Out1 to Out4 and you get a grounded load bridged amp! 😱Nifty👍
Here's a simulation showing this with two input signals, 1kHz and the other channel 2kHz. In this simulation generating an FFT of each output shows no cross-modulation, showing the two amps give independent output signals from a single floating power supply.
So it is possible to run 2 or more channels off a single floating supply using this method - which requires all inputs other than the first one to be isolated and all loads apart from the first to be isolated. In Post 235 I used a small 1:1 transformer for the isolation in my demo.
In this sim (attached) the LHS input is grounded via R3 and link. The LHS load 'Out1' is also grounded.
The RHS input (-) is floating since the link for R12 is open and the RHS load is fully floating.
The power rails are floating using a single supply.
The LHS amp uses the divider R1 and R2 for DC centering. The RHS amp uses the divider R4 and R7 for it's DC centering. Therefore when the supply V3 is shifted relative to common by the LHS amp, then the RHS amps DC reference voltage 'b2' maintains its DC reference point.
So the RHS amp is not affected by the LHS amps output voltage changes.
Notice the top table of harmonics does not contain much 2nd harmonic; you might expect to get some from the other channel running at 2kHz.
But both tables have exactly the same 2nd harmonic levels (which probably comes from numerical noise).
I hope this helps to understand how/why the method I got to work in Post 235 actually works.
Further, it allows bridging of 2 channels, as I showed on the bench.
To bridge in this simulation just use "-Vpk" for the input signal to the RHS channel and make both signals the same frequency. Then connect a single load from Out1 to Out4 and you get a grounded load bridged amp! 😱Nifty👍
My previous tests like Post 226 used PZT222A/2907A buffer drivers. Here I sim the OPA1656 opamp instead that can deliver 75mA. It also adds a gain of 11 for 20dB lower THD (0.02% at 1W, 0.08% at 80W). R11 is increased to 5 Ohms to prevent oscillation with the 20dB higher loop gain. Rin is now 2k (was 1k).
The opamp is modelled using the Level_3a library subcircuit with the parameters: Avol=30Meg GBW=30Meg Slew=10Meg ilimit=75m phimargin=45. For LT-IV the subcicuit name needs to be changed to Level.3a
The opamp is modelled using the Level_3a library subcircuit with the parameters: Avol=30Meg GBW=30Meg Slew=10Meg ilimit=75m phimargin=45. For LT-IV the subcicuit name needs to be changed to Level.3a
An Opps at the end of Post 251 where I said the bridge has outputs from Out1 to Out4.
I meant Out1 to Out3.
I meant Out1 to Out3.
Bench tests with the OPA1656 100mA opamp to check temperature rise of the opamp and stability when adding some open loop gain with the opamp.
I found the temperature rise with a single opamp of the OPA1656 in the SOIC-8 package was about 40C but nearly twice when driven hard into clip. So I tried 2 opamps in parallel (in one SOIC-8 package) tested on a breadboard with a SOIC-8 to DIP8 adapter, and the temperature rise was halved to just under 40C rise with clipping. This looks safe with a PCB board operating at say 50C worst case (ambient temp to 30C). The circuit below is planned for my next PCB with 2 slices in parallel for 4 ohms 200W using opamp drive and to test bridging of 2 PCBs for around 400W into 8 ohms.
I also found the maximum gain that can be added to the open loop gain was around 5 before oscillation at a few MHz starts (with R9 limited to 5 ohms).
I decided to limit the driver/input resistance to 5 ohms (10 ohms per opamp) since this resistor attenuates the loop gain from DC. With 5 ohms the attenuation gain loss is about 3dB. If the opamp gain is more than 5 then the series resistor needs to be higher which drops the maximum net gain back to around 5, so there is no benefit increasing the opamps gain above 5 when there are no compensation capacitors.
I have chosen to not use any compensation for now to keep the design simpler an more reliable for beginners.
Also I will also try single ES3BB-13-F 3A diodes (rather than 2x3A diodes in my earlier tests).
I found the temperature rise with a single opamp of the OPA1656 in the SOIC-8 package was about 40C but nearly twice when driven hard into clip. So I tried 2 opamps in parallel (in one SOIC-8 package) tested on a breadboard with a SOIC-8 to DIP8 adapter, and the temperature rise was halved to just under 40C rise with clipping. This looks safe with a PCB board operating at say 50C worst case (ambient temp to 30C). The circuit below is planned for my next PCB with 2 slices in parallel for 4 ohms 200W using opamp drive and to test bridging of 2 PCBs for around 400W into 8 ohms.
I also found the maximum gain that can be added to the open loop gain was around 5 before oscillation at a few MHz starts (with R9 limited to 5 ohms).
I decided to limit the driver/input resistance to 5 ohms (10 ohms per opamp) since this resistor attenuates the loop gain from DC. With 5 ohms the attenuation gain loss is about 3dB. If the opamp gain is more than 5 then the series resistor needs to be higher which drops the maximum net gain back to around 5, so there is no benefit increasing the opamps gain above 5 when there are no compensation capacitors.
I have chosen to not use any compensation for now to keep the design simpler an more reliable for beginners.
Also I will also try single ES3BB-13-F 3A diodes (rather than 2x3A diodes in my earlier tests).
Hi mr_jj,
Thanks for that link. That's interesting back in 2010 Kenpeter had a resistor across the sensing diodes between the emitters of the power transistors to make it non-switching Class-AB! My hat is off to Kenpeter👍. And HMKD in Post 11 says it was patented in China. I'd not followed that thread back then. Nor had I spotted that thread since I started my thread. Nice find.
Looking further in that 2010 thread I see the Dadson's circuit in Post 44 here (from his AES Oct 1980 article). It shows autobias sensing diodes and the Rush spreader arrangement with Darlington power transistors (copied below for convenience) but notice no resistor across the sensing diodes. I'm not sure if the resistor R1 can do the same as a resistor across the diodes to make it non-switching Class-AB. I haven't sim'd it to see. But interesting to see this arrangement as far back as 1980 - yet sadly it's not mentioned in any of the Power Amp design books.
Thanks for that link. That's interesting back in 2010 Kenpeter had a resistor across the sensing diodes between the emitters of the power transistors to make it non-switching Class-AB! My hat is off to Kenpeter👍. And HMKD in Post 11 says it was patented in China. I'd not followed that thread back then. Nor had I spotted that thread since I started my thread. Nice find.
Looking further in that 2010 thread I see the Dadson's circuit in Post 44 here (from his AES Oct 1980 article). It shows autobias sensing diodes and the Rush spreader arrangement with Darlington power transistors (copied below for convenience) but notice no resistor across the sensing diodes. I'm not sure if the resistor R1 can do the same as a resistor across the diodes to make it non-switching Class-AB. I haven't sim'd it to see. But interesting to see this arrangement as far back as 1980 - yet sadly it's not mentioned in any of the Power Amp design books.
I've been exploring driving the previous version with a level shifter. A level shifter allows bridging without the transformer for isolation as in my Post 235.
I tried a Howland pump using a high voltage 150V opamp OPA455 but the simulation suggested it was only good to a few kHz! So I tried a simple BJT as a current source and it gave enough bandwidth (300kHz) and was linear enough to be used without global feedback.
The BJT (Q2) is referenced to the negative rail of the power amp and needs to supply +/-0.75mA into the amps input for full output swing. Feedback was via R6 and is removed since Out2 AC is now referenced to COM. Feedback is via R9 since Out1 now has the output signal.
With this level shifter the amp still has the same bandwidthof 100kHz.👍.
The level shifter drive and biasing can be worked out later. But first to check it in bridge - as proof of concept - without a transformer for input isolation:
It seems to work OK here but here its only for 16 ohms. For 8 ohms with an 85V supply it needs 2 parallel slices per side (like the previous post).
And it can do two bridges off the same supply, or other PA's😉, since the supply is not floating thanks to the level shifters.
I tried a Howland pump using a high voltage 150V opamp OPA455 but the simulation suggested it was only good to a few kHz! So I tried a simple BJT as a current source and it gave enough bandwidth (300kHz) and was linear enough to be used without global feedback.
The BJT (Q2) is referenced to the negative rail of the power amp and needs to supply +/-0.75mA into the amps input for full output swing. Feedback was via R6 and is removed since Out2 AC is now referenced to COM. Feedback is via R9 since Out1 now has the output signal.
With this level shifter the amp still has the same bandwidthof 100kHz.👍.
The level shifter drive and biasing can be worked out later. But first to check it in bridge - as proof of concept - without a transformer for input isolation:
It seems to work OK here but here its only for 16 ohms. For 8 ohms with an 85V supply it needs 2 parallel slices per side (like the previous post).
And it can do two bridges off the same supply, or other PA's😉, since the supply is not floating thanks to the level shifters.
For the previous post the level shifter drive and biasing can be done as follows:
A Norton style input stage (like the LM3900) with a folded cascode for the positive rail. The folded cascode cancels the 2nd harmonic of Q2 giving 0.01% THD at 1W versus 0.05% without it. If you prefer dominant 2nd then Q8 can be omitted.
Power supply rejection is trimmable with R30 to about -60dB at 100Hz in this sim. Output offset is trimmable with R6.
Gain is 24 (1.7Vpk for 40V out).
A Norton style input stage (like the LM3900) with a folded cascode for the positive rail. The folded cascode cancels the 2nd harmonic of Q2 giving 0.01% THD at 1W versus 0.05% without it. If you prefer dominant 2nd then Q8 can be omitted.
Power supply rejection is trimmable with R30 to about -60dB at 100Hz in this sim. Output offset is trimmable with R6.
Gain is 24 (1.7Vpk for 40V out).
The level shifter needs two high voltage transistors but high gain (HFE) and high Early voltage ones are getting harder to find as good older ones are being dropped from production lines. Cascoding a high beta lower voltage transistor with a low beta high voltage transistor is a good way around this as follows:
Im using 2N5401 and 2N5550 for the high voltage ones (since they are the only high voltage ones in my library at present) but there are many other options still available for a 100V rail (which can rising to 120V with ocassiona higher mains voltages).
One concern is the loss of output swing with cascoding but 1V across the inner transistor appears sufficient with BC847/BC857s losing only about 1V of output swing compared to the previous version.
With cascoding there does not appear to be a need for the folded cascode and no need for PSR trim in the previous post.
Im using 2N5401 and 2N5550 for the high voltage ones (since they are the only high voltage ones in my library at present) but there are many other options still available for a 100V rail (which can rising to 120V with ocassiona higher mains voltages).
One concern is the loss of output swing with cascoding but 1V across the inner transistor appears sufficient with BC847/BC857s losing only about 1V of output swing compared to the previous version.
With cascoding there does not appear to be a need for the folded cascode and no need for PSR trim in the previous post.
Ian, the ZTX757 (PNP) is rated 300V and when I measure its Early voltage, I get 407 volts. You can download its SPICE model parameters from the manufacturer's (Diodes Inc.) website. No need to trust random internet filedumps for model values.
The ZTX857 (NPN) is also rated 300V and when I measure its Early voltage, I get 544 volts. You can download its SPICE model parameters from the manufacturer's website.
Here are I-V curves which I measured myself, in my spare bedroom using my own equipment, for the particular base current which gives 8 milliamps of collector current. Notice (a) the lack of quasi-saturation; (b) the flatness (high Vearly) of the collector current curves. Horizontal axis is |VCE| in volts ; Vertical axis is |ICE| in milliamps. The dam thing plots PNPs in the first quadrant, not the third quadrant. Sorry!
_
The ZTX857 (NPN) is also rated 300V and when I measure its Early voltage, I get 544 volts. You can download its SPICE model parameters from the manufacturer's website.
Here are I-V curves which I measured myself, in my spare bedroom using my own equipment, for the particular base current which gives 8 milliamps of collector current. Notice (a) the lack of quasi-saturation; (b) the flatness (high Vearly) of the collector current curves. Horizontal axis is |VCE| in volts ; Vertical axis is |ICE| in milliamps. The dam thing plots PNPs in the first quadrant, not the third quadrant. Sorry!
_
