hello dear friends,
like you, ages ago, I've been fascinated by the QUAD 405 and the marketing slogans that came with : feedforward compensation, Class B operation, distorsion reduction.
Recently, I considered buying one, being curious about it. Then I got stopped. Why ? Because, surfing on the web, I got the opportunity to downlad info about the QUAD 405, mainly here : Quad 405 Information Page
The main resons why I got stopped are :
1. The sales brochure says this amp is a "current dumping" amp, which is a strange way to describe fine electronics needing to deal with delicate currents supposed to please our ears. What a strange marketing.
2. The schematic looks ugly and complicated. And there is a LM301 at the input, which is outdated now.
But the QUAD 405 always stayed in my mind, since. It is only recently that I realized why this gear was occupying my mind : I still wanted to test the strange ideas of the QUAD 405, but in another context, using modern semiconductors, and possibly, using a full complementary symmetric approach.
I'm presenting here a design idea. I don't expect audiophiles to be seducted by this work because :
1) there is a AD844 at the input (and surely within a few years it will be outdated)
2) there are vertical MOSFETs in the output stage (not recommended for audio)
3) the MOSFETs are underbiased - they operate in Class B.
But, there may exist a category of people wanting to hear what's coming out a redesigned QUAD 405, full complementary symmetric and fitted with MOSFETs at the output.
Of course, one may use lateral MOSFETs. But my simulator only provides me models for vertical MOSFETs. One can however try editing the simulation model of the IRF520 / IRF950, like setting the threshold parameter at 0.8 Volt.
The power supply voltages seem weird : +61V/-64V. Those are the values needed for swinging a 2 Ohm load +40V/-40V without distorsion, near clipping. The output current is thus 10 Amp per MOS. All figures coming from the simulator, of course.
On paper, I think that this circuit exhibits an great quality/complexity compromize. The ouput power is far more than a TDA7294 or LM3886, especially on a 2 Ohm load.
The materialization of such QUAD 405 "revisited" can be in line with ChipAMPS based on TDA7293 and LM3886: a small PCB containing the 4 MOSFETS (like they were a big IC), to be bolted on a big heatsink. Individual isolations needed, of course, unless you use isolated vertical MOSFETs.
Now, let us look from another angle. You'll see the big surprise that will emerge.
Let us say we want to design a CheapAMP, easy to build good for 400 Watt amp on a 2 Ohm load. No doubt we are going to use inexpensive vertical MOSFETs (IRF520/IRF9520), anticipating audiophile people remarks like "they are not suited for audio". That's the foundation of CheapAMP. If you want quality, use lateral MOSFETs (hopefully not faked), and expect to pay a fat premium.
Then we'll scratch our heads about properly biasing the vertical MOSFETs, in a cheap way. Some argue they should be biased at 450mA minimum, for avoiding thermal runaway. Others set them as low as 25mA, expecting the thermal runaway to occur, and stabilize somewhere between 450mA and 550mA when becoming warms. A safety net would be to mount the Vbe multiplier on the heatsink, for a getting a quicker and stronger thermal regulation, but this is far from being cheap. It adds complexity. And, naturally, do you expect the biasing to stay accurate when the amp warms up ? Look the conventional bisaing arrangements : they never compensate for the voltage drop that develops on the source resistor (0.22 ohm to 0.47 ohm). They never compensate for the other transistor increased gate drive. This means, in plain words, that if your thermal management is over-compensating, the more power you are asking, the deeper you will be sitting in Class B, without any overlap current anymore, thus with huge internal crossover distorsion, that the negative feedback God is supposed to cure. Class AB is dirty in essence, and "solid thermal management" may render it even more dirty ! Everybody knows, but nobody publicly admits. Are you listening to Class A amplifiers on a daily base ? Then came the Sanken STD03N/STD03P complementary Darlingtons, fitted with polarization diodes, supposed to instantaneously track the Vbe thresholds of the main transistors. Seemed a good idea on paper (who got the proof they were able to track at 10 KHz ?), but were are they now ? Can you order them through Farnell or Digi-key ? So in essence, Class AB is dirty, and we must know that within the operating range of powers, surely there will be crossover distorsion generated internally, that the negative feedback will be forced to battle.
Then we come to the very annoying part of DIY amplifiers : the input stage. We want a JFET input for getting rid of the input capacitor. Okay, but that's extremely expensive nowadays, as the double JFETs like IFN146 (equivalent 2SK146), 2N3954, 2N3955, 2N3956 have now vanished. Nobody stocks them, and if you can find some, they are crazy expensive like from 7.00 eur to 15.00 eur each. And you need one per channel.
With the widely used OPA-2134 and other FET audio op-amps, one shall feel idiot, building a DIY amplifier, that requires an infamius capacitor at the input.
Then you come to the valid conclusion that, today in 2010, any DIY amp needs to have a JFET audio op-amp at the input like a OPA-2134.
Well, try doing this ... but I warn you : there are not many schemes that are working. Most of the schemas that are working, are using the audio opamps in a current mode, avoiding them to swing a significant output voltage, as a workaround for the dv/dt limitation and the bandwidth of all audio opamps. The opamps are capable of 10 times better performance, but due to the fact that some applications use them as unity gain followers, they need to be stable with total feedback, and thus, there is no other choice than design them as high-gain integrators, adding a big internal compensation capacitor, instead as high-speed amplifiers.
In the past, they were some de-compensated opamps, that needed to be used with a gain greater than 4. One may regret that there are non de-compensated OPA-2134 on the market, featuring a 250V/µs slew-rate and possibly, a gain-bandwith product of 80 MHz. Those ones, surely, could easily act as front-ends in audio power amplifiers. But they have one major disadvantage : they don't exist.
So, now look at the QUAD 405 revisited schematic. Don't you think it is wise to let operate the MOSFETs in Class B, since the start, even at low power. Don't you think we are less cheating, with a strategy like this ? And now look the way that cross-over distorsion gets now cancelled : it occurs using a parallel channel, completely bypassing the MOSFETs. This channel R37, the 50 ohm resistor, directly connected at the output of the high-speed driver. And look : the 3µH coil (L30) is isolating the output from the high-frequency components of the crossover distorsion. And look even further : with R36 (500 ohm), we get the high speed driver completely aware of the crossover distorsion, so he can send the correction signal even before the outputs gets the corrupt signal (delay effect provided by L30). And, this scheme is intrinsically stable, as C32 (120pF) turns the high-speed driver, into a high-speed integrator, indeed.
You will then ask me "what is the idle current of the driver" ? Well, it has been chosen quite low, less than 20mA (T13, T23). This value is compatible with the idea that the driver must be able to deliver the compensation current to the load, using R37 (50 ohm). We can't say it is an extra-cold driver. The dissipation on T13 and T23 is going to be 1.2 Watt. They thus require a small heatsink.
Actually, with this particular schematic relying on an IC as front-end, the driver is actually capable of delivering a lot of current. Look how the AD844 gets connected to the driver. At idle, the current of the driver is defined by the idle current of the AD844, multiplied by a certain factor because of the ratio between R10 and R56. But imagine now that the AD844 is delivering 2 Volt at pin 6, which is loaded by R33 (82 ohm). The current deliverd by pin 6 is 24mA. The current adds to the idle current of the AD844, and gets conducted by T10 and R10. Those 24mA get multiplied by a certain factor by T12/T13. So, in essence, the driver may be able to deliver something close to 200mA. You now understand that, unlike conventional audio-amp drivers, this particular driver is able to deliver far more than twice its idle current.
But, isn't a peak driver current of 200mA a little bit excessive ? Yes ! Therefore I have adeed the limiter, consisting of T30/T31. One can play with the value of R33, defining the threshold. The idea is to avoid any latch-up or catastrophic behaviour when the amp is saturating, being a voltage saturation (asking 80V at the output, when there is no supply for this), or, more tricky, when dynamic saturation occurs like when there is an excessive dV/dt at the input.
I have added R13 C10 R23 C20 for enabling a AC gain that's greater than the DC gain, at the driver level. One will see that the THD increases a little bit when chosing 10K for R13 and R23. On the other side, when chosing a very small value for R13 and R23 (like 3,3 ohm), the open loop gain increases, but I see no more THD decrease at 1Khz. Maybe the optimal value for R13 and R23 is the value where THD ceases to significantly drop ? Now that we have chosen a value of 18 ohm for R13 and R23, C10 and C20 need to be 470 µF if we want them to become neglectible above a frequency of 20 Hz. But, are we sure we need any open-loop gain boost at frequencies as low as 20 Hz ? I don't think so, so I have selected 220 µF for C10 and C20.
Why so big capacitors in parallel with the Vbe multipliers ? I wanted their impedance to be very small against R37, the resistor that feeds the correction signal. Please note the MOSFETs get underbiased, using those Vbe multipliers. The bias voltage is significantly lower than the MOSFETs threshold voltages. There is no need for mounting the Vbe multipliers on the heatsink. This is pure Class B, here.
One may mount the Vbe multipliers on the heatsink, and adjust them for getting a small idle current in the MOSFETs, something between 50mA and 500mA. I have not investigated yet. Sort of QUAD 405 hybridation.
Hey, but wait a minute, there is no input capacitor in your circuit, and the AD844 doesn't have JFETs at the input ! I know. The AD844 is full complementary from A to Z. His input bias currents are not the bias usually needed to feed the base currents of an internal long tailed pair. His bias currents are the difference between the bias requirements of two complementary functional blocks. Those those blocks are laser-trimmed at the factory. Look the schematic of pin 3. The offset voltage and input bias currents of the AD844 are laser-trimmed to minimize DC errors. Vos drift is typically 1µV/°C and bias current drift is typically 9nA/°C. This makes the AD844 compete in a different league than mainstream bipolar opamps, and this is the reason why I have omitted the input capacitor on the schematic. If you use a low-value volume pot, like 10K, you should experience no scratch. And if you use a modern silicon volume pot (CS3310, CS3318 and others), there is no risk of the very tiny polarization current causing electromigration or microdiodes on the potentiometer tap.
So, in total, how much "capacitor sound" in this design ?
Maybe zero ... as the capacitors (C10, C11, C20, C21) are all included inside the negative feedback loop.
So, what is thus the surprise ?
The QUAD 405 structure enables cheap and compact amplifiers, with decent quality, however not audiophile grade.
The QUAD 405 provides a scientific, objectivable and reliable answer to crossover distorsion, that no conventional dumb Class AB amplifier can provide.
As usual, I'm attaching the Tina T.I. 7 schematic ready for simulation.
Voilà, that's all I have to say - for the moment.
Cheers,
Steph
like you, ages ago, I've been fascinated by the QUAD 405 and the marketing slogans that came with : feedforward compensation, Class B operation, distorsion reduction.
Recently, I considered buying one, being curious about it. Then I got stopped. Why ? Because, surfing on the web, I got the opportunity to downlad info about the QUAD 405, mainly here : Quad 405 Information Page
The main resons why I got stopped are :
1. The sales brochure says this amp is a "current dumping" amp, which is a strange way to describe fine electronics needing to deal with delicate currents supposed to please our ears. What a strange marketing.
2. The schematic looks ugly and complicated. And there is a LM301 at the input, which is outdated now.
But the QUAD 405 always stayed in my mind, since. It is only recently that I realized why this gear was occupying my mind : I still wanted to test the strange ideas of the QUAD 405, but in another context, using modern semiconductors, and possibly, using a full complementary symmetric approach.
I'm presenting here a design idea. I don't expect audiophiles to be seducted by this work because :
1) there is a AD844 at the input (and surely within a few years it will be outdated)
2) there are vertical MOSFETs in the output stage (not recommended for audio)
3) the MOSFETs are underbiased - they operate in Class B.
But, there may exist a category of people wanting to hear what's coming out a redesigned QUAD 405, full complementary symmetric and fitted with MOSFETs at the output.
Of course, one may use lateral MOSFETs. But my simulator only provides me models for vertical MOSFETs. One can however try editing the simulation model of the IRF520 / IRF950, like setting the threshold parameter at 0.8 Volt.
The power supply voltages seem weird : +61V/-64V. Those are the values needed for swinging a 2 Ohm load +40V/-40V without distorsion, near clipping. The output current is thus 10 Amp per MOS. All figures coming from the simulator, of course.
On paper, I think that this circuit exhibits an great quality/complexity compromize. The ouput power is far more than a TDA7294 or LM3886, especially on a 2 Ohm load.
The materialization of such QUAD 405 "revisited" can be in line with ChipAMPS based on TDA7293 and LM3886: a small PCB containing the 4 MOSFETS (like they were a big IC), to be bolted on a big heatsink. Individual isolations needed, of course, unless you use isolated vertical MOSFETs.
Now, let us look from another angle. You'll see the big surprise that will emerge.
Let us say we want to design a CheapAMP, easy to build good for 400 Watt amp on a 2 Ohm load. No doubt we are going to use inexpensive vertical MOSFETs (IRF520/IRF9520), anticipating audiophile people remarks like "they are not suited for audio". That's the foundation of CheapAMP. If you want quality, use lateral MOSFETs (hopefully not faked), and expect to pay a fat premium.
Then we'll scratch our heads about properly biasing the vertical MOSFETs, in a cheap way. Some argue they should be biased at 450mA minimum, for avoiding thermal runaway. Others set them as low as 25mA, expecting the thermal runaway to occur, and stabilize somewhere between 450mA and 550mA when becoming warms. A safety net would be to mount the Vbe multiplier on the heatsink, for a getting a quicker and stronger thermal regulation, but this is far from being cheap. It adds complexity. And, naturally, do you expect the biasing to stay accurate when the amp warms up ? Look the conventional bisaing arrangements : they never compensate for the voltage drop that develops on the source resistor (0.22 ohm to 0.47 ohm). They never compensate for the other transistor increased gate drive. This means, in plain words, that if your thermal management is over-compensating, the more power you are asking, the deeper you will be sitting in Class B, without any overlap current anymore, thus with huge internal crossover distorsion, that the negative feedback God is supposed to cure. Class AB is dirty in essence, and "solid thermal management" may render it even more dirty ! Everybody knows, but nobody publicly admits. Are you listening to Class A amplifiers on a daily base ? Then came the Sanken STD03N/STD03P complementary Darlingtons, fitted with polarization diodes, supposed to instantaneously track the Vbe thresholds of the main transistors. Seemed a good idea on paper (who got the proof they were able to track at 10 KHz ?), but were are they now ? Can you order them through Farnell or Digi-key ? So in essence, Class AB is dirty, and we must know that within the operating range of powers, surely there will be crossover distorsion generated internally, that the negative feedback will be forced to battle.
Then we come to the very annoying part of DIY amplifiers : the input stage. We want a JFET input for getting rid of the input capacitor. Okay, but that's extremely expensive nowadays, as the double JFETs like IFN146 (equivalent 2SK146), 2N3954, 2N3955, 2N3956 have now vanished. Nobody stocks them, and if you can find some, they are crazy expensive like from 7.00 eur to 15.00 eur each. And you need one per channel.
With the widely used OPA-2134 and other FET audio op-amps, one shall feel idiot, building a DIY amplifier, that requires an infamius capacitor at the input.
Then you come to the valid conclusion that, today in 2010, any DIY amp needs to have a JFET audio op-amp at the input like a OPA-2134.
Well, try doing this ... but I warn you : there are not many schemes that are working. Most of the schemas that are working, are using the audio opamps in a current mode, avoiding them to swing a significant output voltage, as a workaround for the dv/dt limitation and the bandwidth of all audio opamps. The opamps are capable of 10 times better performance, but due to the fact that some applications use them as unity gain followers, they need to be stable with total feedback, and thus, there is no other choice than design them as high-gain integrators, adding a big internal compensation capacitor, instead as high-speed amplifiers.
In the past, they were some de-compensated opamps, that needed to be used with a gain greater than 4. One may regret that there are non de-compensated OPA-2134 on the market, featuring a 250V/µs slew-rate and possibly, a gain-bandwith product of 80 MHz. Those ones, surely, could easily act as front-ends in audio power amplifiers. But they have one major disadvantage : they don't exist.
So, now look at the QUAD 405 revisited schematic. Don't you think it is wise to let operate the MOSFETs in Class B, since the start, even at low power. Don't you think we are less cheating, with a strategy like this ? And now look the way that cross-over distorsion gets now cancelled : it occurs using a parallel channel, completely bypassing the MOSFETs. This channel R37, the 50 ohm resistor, directly connected at the output of the high-speed driver. And look : the 3µH coil (L30) is isolating the output from the high-frequency components of the crossover distorsion. And look even further : with R36 (500 ohm), we get the high speed driver completely aware of the crossover distorsion, so he can send the correction signal even before the outputs gets the corrupt signal (delay effect provided by L30). And, this scheme is intrinsically stable, as C32 (120pF) turns the high-speed driver, into a high-speed integrator, indeed.
You will then ask me "what is the idle current of the driver" ? Well, it has been chosen quite low, less than 20mA (T13, T23). This value is compatible with the idea that the driver must be able to deliver the compensation current to the load, using R37 (50 ohm). We can't say it is an extra-cold driver. The dissipation on T13 and T23 is going to be 1.2 Watt. They thus require a small heatsink.
Actually, with this particular schematic relying on an IC as front-end, the driver is actually capable of delivering a lot of current. Look how the AD844 gets connected to the driver. At idle, the current of the driver is defined by the idle current of the AD844, multiplied by a certain factor because of the ratio between R10 and R56. But imagine now that the AD844 is delivering 2 Volt at pin 6, which is loaded by R33 (82 ohm). The current deliverd by pin 6 is 24mA. The current adds to the idle current of the AD844, and gets conducted by T10 and R10. Those 24mA get multiplied by a certain factor by T12/T13. So, in essence, the driver may be able to deliver something close to 200mA. You now understand that, unlike conventional audio-amp drivers, this particular driver is able to deliver far more than twice its idle current.
But, isn't a peak driver current of 200mA a little bit excessive ? Yes ! Therefore I have adeed the limiter, consisting of T30/T31. One can play with the value of R33, defining the threshold. The idea is to avoid any latch-up or catastrophic behaviour when the amp is saturating, being a voltage saturation (asking 80V at the output, when there is no supply for this), or, more tricky, when dynamic saturation occurs like when there is an excessive dV/dt at the input.
I have added R13 C10 R23 C20 for enabling a AC gain that's greater than the DC gain, at the driver level. One will see that the THD increases a little bit when chosing 10K for R13 and R23. On the other side, when chosing a very small value for R13 and R23 (like 3,3 ohm), the open loop gain increases, but I see no more THD decrease at 1Khz. Maybe the optimal value for R13 and R23 is the value where THD ceases to significantly drop ? Now that we have chosen a value of 18 ohm for R13 and R23, C10 and C20 need to be 470 µF if we want them to become neglectible above a frequency of 20 Hz. But, are we sure we need any open-loop gain boost at frequencies as low as 20 Hz ? I don't think so, so I have selected 220 µF for C10 and C20.
Why so big capacitors in parallel with the Vbe multipliers ? I wanted their impedance to be very small against R37, the resistor that feeds the correction signal. Please note the MOSFETs get underbiased, using those Vbe multipliers. The bias voltage is significantly lower than the MOSFETs threshold voltages. There is no need for mounting the Vbe multipliers on the heatsink. This is pure Class B, here.
One may mount the Vbe multipliers on the heatsink, and adjust them for getting a small idle current in the MOSFETs, something between 50mA and 500mA. I have not investigated yet. Sort of QUAD 405 hybridation.
Hey, but wait a minute, there is no input capacitor in your circuit, and the AD844 doesn't have JFETs at the input ! I know. The AD844 is full complementary from A to Z. His input bias currents are not the bias usually needed to feed the base currents of an internal long tailed pair. His bias currents are the difference between the bias requirements of two complementary functional blocks. Those those blocks are laser-trimmed at the factory. Look the schematic of pin 3. The offset voltage and input bias currents of the AD844 are laser-trimmed to minimize DC errors. Vos drift is typically 1µV/°C and bias current drift is typically 9nA/°C. This makes the AD844 compete in a different league than mainstream bipolar opamps, and this is the reason why I have omitted the input capacitor on the schematic. If you use a low-value volume pot, like 10K, you should experience no scratch. And if you use a modern silicon volume pot (CS3310, CS3318 and others), there is no risk of the very tiny polarization current causing electromigration or microdiodes on the potentiometer tap.
So, in total, how much "capacitor sound" in this design ?
Maybe zero ... as the capacitors (C10, C11, C20, C21) are all included inside the negative feedback loop.
So, what is thus the surprise ?
The QUAD 405 structure enables cheap and compact amplifiers, with decent quality, however not audiophile grade.
The QUAD 405 provides a scientific, objectivable and reliable answer to crossover distorsion, that no conventional dumb Class AB amplifier can provide.
As usual, I'm attaching the Tina T.I. 7 schematic ready for simulation.
Voilà, that's all I have to say - for the moment.
Cheers,
Steph
Attachments
Last edited:
here is v0.2
In QUAD 405 theory, the block gain is supposed to be infinite ... and we really need this otherwise THD at low level is terrible. The v0.1 schematic had a wise, moderate approach regarding open-loop gain, but this is a wrong approach when dealing with a QUAD 405 arrangement. The v0.1 schematic had 0.275% THD with an input of 20mV 1KHz, output loaded with 2 ohm.
In v0.2, the open-loop gain is now set and optimized, not using a 80Vpp output wave, but using a 800mVpp output wave (20mV 1KHz at the input).
The open-loop gain got increased using three adjustments :
- R34 at pin 5 of the AD844 (voltage gain section)
- R33 at pin 5 of the AD844 (output stage, defining the AD844 audio current)
- R13 and R23 in the driver
With v0.2, the THD at low level with 20mV 1KHz input is only 0.025 % in a 2 ohm load.
Seeing this vast improvement, I double checked that the MOSFETs are still underbiased.
THD with 1000mV 1KHz input, leading to a 80Vpp swing, is 0.025% in a 2 Ohm load.
THD with 1000mV 1KHz input, leading to a 80Vpp swing, is 0.014% in a 4 Ohm load.
THD with 1000mV 1KHz input, leading to a 80Vpp swing, is 0.009% in a 8 Ohm load.
And I double checked that the MOSFETs are still underbiased.
This v0.2 revision brings a tweak in the Q405 feedback network : Z1, Z2, Z3 and Z4.
I have played with some values, following the Q405 rule Z2 / Z3 = Z1 / Z4. There is not so much degree in freedom in this, in the real world.
Z3 : this capacitor value must be an order of magnitude bigger than the surrounding parasitic capacitances.
Z2 : one would like it to be higher than 500 ohm, but in order to keep the voltage amplification at x40, R35 must be increased accordingly. However, it seems that the AD844 likes low resistor values on pin 2 (R35).
Z1 : maybe using a value in the order of 1,0 µH to 3 µH, is it possible to get rid of the 2nd coil, used to isolate capacitive loads. Need to investigate this. Reducing component count is always good.
Z4 : intuitively, this can't be lower than 25 ohm, because it then loads the driver a lot for generating the compensation current.
During the optimization process of the open-loop gain I discovered that there is a practical limit to the maximum open-loop gain that is allowed. If the open-loop gain is too high, the simulator takes more time than usual to compute the 500mV 10 KHz square wave 200µs Transient Analysis, which is a strong indication that a real-world circuit will experience tiny instabilities. The open-loop gain that I have chosen in v0.2 allows the simulator to compute the 500mV 10 KHz square wave 200µs Transient Analysis in a smooth way, without long pauses, but the computation is still slower than with v0.1. This needs to be remembered in case of v0.2 instabilities in the real-world.
I have not yet measured the impact of a capacitive load.
Cheers,
Steph
In QUAD 405 theory, the block gain is supposed to be infinite ... and we really need this otherwise THD at low level is terrible. The v0.1 schematic had a wise, moderate approach regarding open-loop gain, but this is a wrong approach when dealing with a QUAD 405 arrangement. The v0.1 schematic had 0.275% THD with an input of 20mV 1KHz, output loaded with 2 ohm.
In v0.2, the open-loop gain is now set and optimized, not using a 80Vpp output wave, but using a 800mVpp output wave (20mV 1KHz at the input).
The open-loop gain got increased using three adjustments :
- R34 at pin 5 of the AD844 (voltage gain section)
- R33 at pin 5 of the AD844 (output stage, defining the AD844 audio current)
- R13 and R23 in the driver
With v0.2, the THD at low level with 20mV 1KHz input is only 0.025 % in a 2 ohm load.
Seeing this vast improvement, I double checked that the MOSFETs are still underbiased.
THD with 1000mV 1KHz input, leading to a 80Vpp swing, is 0.025% in a 2 Ohm load.
THD with 1000mV 1KHz input, leading to a 80Vpp swing, is 0.014% in a 4 Ohm load.
THD with 1000mV 1KHz input, leading to a 80Vpp swing, is 0.009% in a 8 Ohm load.
And I double checked that the MOSFETs are still underbiased.
This v0.2 revision brings a tweak in the Q405 feedback network : Z1, Z2, Z3 and Z4.
I have played with some values, following the Q405 rule Z2 / Z3 = Z1 / Z4. There is not so much degree in freedom in this, in the real world.
Z3 : this capacitor value must be an order of magnitude bigger than the surrounding parasitic capacitances.
Z2 : one would like it to be higher than 500 ohm, but in order to keep the voltage amplification at x40, R35 must be increased accordingly. However, it seems that the AD844 likes low resistor values on pin 2 (R35).
Z1 : maybe using a value in the order of 1,0 µH to 3 µH, is it possible to get rid of the 2nd coil, used to isolate capacitive loads. Need to investigate this. Reducing component count is always good.
Z4 : intuitively, this can't be lower than 25 ohm, because it then loads the driver a lot for generating the compensation current.
During the optimization process of the open-loop gain I discovered that there is a practical limit to the maximum open-loop gain that is allowed. If the open-loop gain is too high, the simulator takes more time than usual to compute the 500mV 10 KHz square wave 200µs Transient Analysis, which is a strong indication that a real-world circuit will experience tiny instabilities. The open-loop gain that I have chosen in v0.2 allows the simulator to compute the 500mV 10 KHz square wave 200µs Transient Analysis in a smooth way, without long pauses, but the computation is still slower than with v0.1. This needs to be remembered in case of v0.2 instabilities in the real-world.
I have not yet measured the impact of a capacitive load.
Cheers,
Steph
Attachments
Last edited:
here is v0.3
Actually, this is v0.2 with 50mA bias current in each MOSFET.
The idea is to enable the AC Analysis / AC transfer Characteristic simulator running with "something different" than the small signal hypothesis exercised inside the Class B dead-zone.
Indeed, biasing each MOSFET with a 50mA quiescent current, we do expect a significant open-loop gain increase. We are thus putting ourselves in the "worst case" small signal analysis, the "worst case" being when the open loop gain gets maximized.
A good design practice is to ensure that, even if the circuit becomes Class AB biased (50mA quiescent current in each MOSFET), that the closed loop remains stable, with no gain peaking at a certain frequency.
Well, dear friends, here is the sad news : v2.0 fails the test.
Revision v3.0 shows that using the exact same open-loop gain as v2.0, it exhibits a strong gain peaking at 7 MHz, which is a strong indication that a real-world circuit will oscillate, if each MOSFET gets a quiescent current of 50 mA.
As a by-product, this finding ruins the possibility of operating the QUAD 405 v0.2 in a multimodal way, with some bias applied for pleasing people not entirely trusting the QUAD 405 feedforward topology.
I need to redesign the v2.0 open-loop gain. I'll do this later on. Time to get some sleep.
Must say I don't feel happy with the big capacitors in the drivers emitters. I don't like the idea of having the driver experiencing an open-loop gain drop at very low frequencies. I'm still in search of a full DC coupling scheme between the AD844 and the BC55x/BD32x driver.
Must say also that using a dumb 330 ohm resistive load (plus diode) is destroying the drive symmetry. The BC55x driver base can be quickly activated, with a voltage drop developing on the 330 ohm resistor. But the de-activation of the BC55x driver only relies on the BC55x base self-discharging in the 330 ohm (plus diode). That's quite slow in comparison with the AD844. Presently, what's concerning the BC55x base discharge, there is no active device in charge of short-circuiting the base.
A current mirror arrangement usually provides the symmetric fast base charge / fast base discharge feature, but how to implement it here, knowing that the the BC55x/BD32x driver must deliver a rock-solid idle curent ?
The BC55x/BD32x driver must provide a rock-solid idle current, because if the idle current is evanescent, variable, the Vbe multipliers will also deliver an evanescent, variable biasing voltage. And nobody, at the end of the day, will be able to tell if the QUAD 405 revisited is now operating in pure Class B, with a huge dead zone, with a moderate dead zone, or in Class AB with a small quiescent current, or in Class AB with a strong quiescent current.
As usual, there is the attached Tina 7 T.I. schematic ready for simulation.
Any suggestion welcome.
Cheers,
Steph
Actually, this is v0.2 with 50mA bias current in each MOSFET.
The idea is to enable the AC Analysis / AC transfer Characteristic simulator running with "something different" than the small signal hypothesis exercised inside the Class B dead-zone.
Indeed, biasing each MOSFET with a 50mA quiescent current, we do expect a significant open-loop gain increase. We are thus putting ourselves in the "worst case" small signal analysis, the "worst case" being when the open loop gain gets maximized.
A good design practice is to ensure that, even if the circuit becomes Class AB biased (50mA quiescent current in each MOSFET), that the closed loop remains stable, with no gain peaking at a certain frequency.
Well, dear friends, here is the sad news : v2.0 fails the test.
Revision v3.0 shows that using the exact same open-loop gain as v2.0, it exhibits a strong gain peaking at 7 MHz, which is a strong indication that a real-world circuit will oscillate, if each MOSFET gets a quiescent current of 50 mA.
As a by-product, this finding ruins the possibility of operating the QUAD 405 v0.2 in a multimodal way, with some bias applied for pleasing people not entirely trusting the QUAD 405 feedforward topology.
I need to redesign the v2.0 open-loop gain. I'll do this later on. Time to get some sleep.
Must say I don't feel happy with the big capacitors in the drivers emitters. I don't like the idea of having the driver experiencing an open-loop gain drop at very low frequencies. I'm still in search of a full DC coupling scheme between the AD844 and the BC55x/BD32x driver.
Must say also that using a dumb 330 ohm resistive load (plus diode) is destroying the drive symmetry. The BC55x driver base can be quickly activated, with a voltage drop developing on the 330 ohm resistor. But the de-activation of the BC55x driver only relies on the BC55x base self-discharging in the 330 ohm (plus diode). That's quite slow in comparison with the AD844. Presently, what's concerning the BC55x base discharge, there is no active device in charge of short-circuiting the base.
A current mirror arrangement usually provides the symmetric fast base charge / fast base discharge feature, but how to implement it here, knowing that the the BC55x/BD32x driver must deliver a rock-solid idle curent ?
The BC55x/BD32x driver must provide a rock-solid idle current, because if the idle current is evanescent, variable, the Vbe multipliers will also deliver an evanescent, variable biasing voltage. And nobody, at the end of the day, will be able to tell if the QUAD 405 revisited is now operating in pure Class B, with a huge dead zone, with a moderate dead zone, or in Class AB with a small quiescent current, or in Class AB with a strong quiescent current.
As usual, there is the attached Tina 7 T.I. schematic ready for simulation.
Any suggestion welcome.
Cheers,
Steph
Attachments
here is v0.4
Actually, this is v0.3 with a frequency compensation network added on pin 5 of the AD844. Beware some mods in the numbering scheme of some components, especially in the limiter.
The circuit looks rock-solid stable.
The 500mV 10 KHz square wave 200µs Transient Analysis now computes very fast.
The distorsion figures are still optimal.
This sounds healthy and reassuring.
As usual, there is the attached Tina 7 T.I. schematic ready for simulation.
Need now to run a full test with Class B operation, the "real" QUAD 405 operation mode.
Cheers,
Steph
Actually, this is v0.3 with a frequency compensation network added on pin 5 of the AD844. Beware some mods in the numbering scheme of some components, especially in the limiter.
The circuit looks rock-solid stable.
The 500mV 10 KHz square wave 200µs Transient Analysis now computes very fast.
The distorsion figures are still optimal.
This sounds healthy and reassuring.
As usual, there is the attached Tina 7 T.I. schematic ready for simulation.
Need now to run a full test with Class B operation, the "real" QUAD 405 operation mode.
Cheers,
Steph
Attachments
here is v0.5
Surprise, surprise : no AD844 anymore. Instead, there is a quick and dirty complementary emitter follower, shorted to ground. This is an inverting amp configuration providing up to 77dB open-loop voltage gain. When the voltage gain gets reduced to 6 dB, the THD reduction is effective, with only 0.033% THD when swinging 80Vpp on a 2 ohm load, and 0.036% THD when swinging 2Vpp on a 2 ohm load. All this with the MOSFETs underbiased, each at 3.30V, instead of 3.72V needed for entering a 50mA conduction.
This is not a proper revised QUAD 405 however, due to the low voltage gain (6 dB) and due to the small input resistance (250 ohm). But it may constitue an universal linearized output stage, good for any amp.
One may think an OPA-134 may suffice as front-end, but this is not true as the 250 ohm resistor needs to be driven 40Vpp for getting a 80Vpp output wave. In other words, one needs to swing + 20V and - 20V with a current of 80mA. It means that another quick and dirty "thing" is still needed between an OPA-134 front-end, and the 250 ohm resistor. Will a certain amount of global feedback (say 20 dB) add some polish to the signal ? We shall see.
As usual, there is the attached Tina 7 T.I. schematic ready for simulation.
Cheers,
Steph
Surprise, surprise : no AD844 anymore. Instead, there is a quick and dirty complementary emitter follower, shorted to ground. This is an inverting amp configuration providing up to 77dB open-loop voltage gain. When the voltage gain gets reduced to 6 dB, the THD reduction is effective, with only 0.033% THD when swinging 80Vpp on a 2 ohm load, and 0.036% THD when swinging 2Vpp on a 2 ohm load. All this with the MOSFETs underbiased, each at 3.30V, instead of 3.72V needed for entering a 50mA conduction.
This is not a proper revised QUAD 405 however, due to the low voltage gain (6 dB) and due to the small input resistance (250 ohm). But it may constitue an universal linearized output stage, good for any amp.
One may think an OPA-134 may suffice as front-end, but this is not true as the 250 ohm resistor needs to be driven 40Vpp for getting a 80Vpp output wave. In other words, one needs to swing + 20V and - 20V with a current of 80mA. It means that another quick and dirty "thing" is still needed between an OPA-134 front-end, and the 250 ohm resistor. Will a certain amount of global feedback (say 20 dB) add some polish to the signal ? We shall see.
As usual, there is the attached Tina 7 T.I. schematic ready for simulation.
Cheers,
Steph
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.