New 220 Watt CMOS monoblock

[Leslie Horn]

Hmmm, overkill?
No protection at all (yet)
as this will be the prototype.

http://www.lightstrip.com/0424/2b-rev1bPRN.gif

It simulates fine, from 11 Hz to 22 Khz, with 2 volts p-p input
and 60 volts p-p output into 8 ohms.

I probably need to add protection diodes later, to prevent breakdown
of the drivers and mosfets should input be overdriven much past 2 v p-p.

Note the drivers are not Q2-Q3-Q25-Q26; they are Q4-q21-q28-q18.

The amplifier is bridged, for a lower supply voltage and to facilitate
lower voltage devices. Note the phase inverter (capacitor coupled)
at the very input. I thought of taking the alternate phase from the
collector of Q14 and feeding it directly to the base of the Q28 driver
stage, but it was far less stable than as it is now.

All right, is this design freakin' crazy enough?

I'm gonna etch the board soon...

[AndrewT]

Hi,
Q1. Are irf540 & 9540 100v devices? Then 54V rails are driving them a bit too hard. This will be exacerbated by bridging because when driving an 8 ohm load each half of the bridge thinks its is driving a 4 ohm load and draws that much current and dissipates that much heat. If you have 6ohm or 4 ohm speakers then the outputs are working even harder.

Q2. Does R15 &16 need to connect to NFB/output?

Q3. Does the Miller compensation need to be so big?

Q4. Have you tried adding emitter degeneration resistors to the LTP pairs?

Q5. Have you calculated R5, 6k2 seems far too big. Same with VR27.

Q6. Are you running the ltp sink @ 3mA? then the voltage across r23 & 24 is a bit high @ 7V try 1k0 instead giving 1v5 and allow R35 to be reduced.
You may need to add a cap // R58.
Q4 is the VAS and Q21 is the current source (sink). Q2 & 3 are the complementary drivers connected as compound follower pairs (Skizlai but with FETS).

[Leslie Horn]

Q1. Are irf540 & 9540 100v devices? Then 54V rails are driving them a bit too hard. This will be exacerbated by bridging because when driving an 8 ohm load each half of the bridge thinks its is driving a 4 ohm load and draws that much current and dissipates that much heat. If you have 6ohm or 4 ohm speakers then the outputs are working even harder.

Yeah, I know that, but I am running class A / AB, and idling them quite a bit above cutoff. So the output swing is only +/- 30 volts per side, for 54 + 30 = 74 volts, max, across the cutoff MOSFETS.

I was tempted to use 200 volt devices, and scrap the bridge idea, and go with 70 volt rails, yada yada, but I had this other idea here

Q2. Does R15 &16 need to connect to NFB/output?

Yes, in this case, I have Q2 and Q3 as level shifters, with no real power gain.

Q3. Does the Miller compensation need to be so big?

Not really, but it seems to get a lower noise floor. I had the caps at 220 pf originally.
Here, with 470 pf, it starts to act as an integrator above 25 khz, and the output is quieter with zero volts RMS input. 390 pF is a compromise.

Q4. Have you tried adding emitter degeneration resistors to the LTP pairs?

Nope. I ought to. But I already laid out the PCB. Manually. I'm sorry I did!

Q5. Have you calculated R5, 6k2 seems far too big. Same with VR27.

Seemed high to me, too, as the current through R5 is less than Q9's base current, but I have seen this relationship in other bias servos. Since Q9 is essentially a Vbe multiplier, I need the extra voltage for the level shifters. Also, I am running this class A up to several watts, so I do have a little more idle bias.

Q6. Are you running the ltp sink @ 3mA? then the voltage across r23 & 24 is a bit high @ 7V try 1k0 instead giving 1v5 and allow R35 to be reduced.
You may need to add a cap // R58.

In practice, the cascoded pair runs close to 6.5ma. it would be 7ma, but the Ic of Q10 is actually the Ic of Q11 minus the sum of the base currents of Q10 and Q11, times two. Q5 and Q14 each get about 2.88 ma at the collectors. I found I could get better linearity by driving the VAP/drivers harder, and that means more open loop gain in the input cascode pairs. I originally tried lower values of collector resistance, and the OL gain was too low to drive the VAS to 60V P-P without distortion. Much lower, and it even became unstable!

Q4 is the VAS and Q21 is the current source (sink). Q2 & 3 are the complementary drivers connected as compound follower pairs (Skizlai but with FETS).

Yes, and no. The emitters of Q4 and Q21 are returned to the drains of Q7,8 and Q1,6, respectively.
They are in essence, drain followers. The original idea for this configuration was to get more output swing (closer to rail-to-rail), as the four volt gate threshold is not following the output, as in a source follower, but it didn't work out that way. In any event, I am driving the heck out of the VAS, quite nonlinearly, I might add. But the output shows 0.05% 2nd harmonic at 1 Khz, at full power. 3rd is down to 0.033% (10mV at 30V Pk output). I ain't complaining, even though I could tweak some and get the distortion lower. Again, this is simulated...

I have to physically build it next!

[amplifierguru]

Hi Leslie,

Some constructive criticism. This is the very type of amplifier that attracted criticism way back for it poor slewing rate due to that HEAVY miller comp on the Vas and no degen.

I would degenerate the input stage emitters by at least 2 times their junction impedance roughly 56 ohms for each mA of emitter current. Then apply a miller Ccomp of maybe 10-20pF to define the stage's rolloff so it's independant of the vagaries of the device. Then I would start with Cf from Vas collector to the input feedback loop base junction (Q13,Q29 bases) . This needs to be trimmed of the order of 47pF and result in much higher slew rates and much lower THD especially at 20KHz.

[AndrewT]

Hi,
If you only intend to run the output stage to 60Vpp then supply these with +-45Vrails. Keep the high voltage (low current) rails for the front end.
Something seems fundamentally wrong with your front end if the VAS can only swing 60Vpp from 108Vrail potential. I don't have the knowledge to help here. Could reducing R35 substantially help?
The reduction in C5 (miller comp) should be a factor of ten, not -30%. Again instability here indicates something wrong.
I have seen it on threads here that the Sziklai pair (better spelling this time) is quite difficult to stabilise with bipolar outputs. I'm guessing from that, it might be even more difficult with faster FETS hung on the end. Could this be another clue to needing that 390p for C5?
Fets should be able to output upto at least 100kHz (at all powers and into about half final impedance (although only short term) while you check for instability. Try hanging a range of caps on the output while you run more simms.