kenpeter said:
Not sure how this would work for the fet version. The current mirrors are the gain stage, what exactly are you replacing them with?
The buffer in front of the centerpoint of Q3/Q4 makes this a voltage-feedback opamp. Otherwise, it's current-feedback. I wanted to design a fairly flexible opamp, not just a headphone driver, so I went for the VFB design.
Also, there is better CMRR and CM distortion due to cancellation between the two identical diamond structures. This is why I included the diodes and resistors in series with the power rails - both buffers are operated at the same Vce and Ic.
nuvistor said:
Thermal stability seems to be good, the bias does not drift much as it warms up. I can set it for 20mA cold and it will definitely be within 15mA-25mA if I check it hot. I have tried bias currents up to 50mA and it doesn't go into runaway. I have also driven 4 ohm loads with it without a problem.
I haven't; it should work fine, but AC performance will suffer since less feedback will be available. It may be possible to mitigate this by lowering the input degen resistors further, which will increase the open-loop gain. In my application I actually run at a gain of -12db in an inverting configuration, with the DC servo output fed into the (+) input (this makes more sense if you know how the rest of the circuit works).
I really think the high order distortion problem was due to the turn-off time of the BJTs. Yes, the slew rate of audio signals is typically low, however you still need to move through the crossover region. It doesn't matter how fast you do so, just that it happens. If the output devices don't turn off exactly when they should, you have transient cross-conduction which shows up as low-level high-order harmonics.
The small signal MOSFETs are Zetex ZVN2106A/ZVP2106A, and the output MOSFETs are Exicon ECX10N20/ECX10P20. There are some details on my website.Lumba Ogir said:
Mr Evil,
those MOSFETs are indeed nice but I would not use them in a subwoofer amplifier because of their inferior current delivering ability.
I would not use them in VAS either as it is a transimpedance amplifier performing current to voltage conversion, which for instance, ideally should have zero input impedance. FETs are by nature transconductance devices, excellent for voltage to current transformation, requiring the opposite principal properties.
Expect high distortion in that stage.
those MOSFETs are indeed nice but I would not use them in a subwoofer amplifier because of their inferior current delivering ability.
I would not use them in VAS either as it is a transimpedance amplifier performing current to voltage conversion, which for instance, ideally should have zero input impedance. FETs are by nature transconductance devices, excellent for voltage to current transformation, requiring the opposite principal properties.
Expect high distortion in that stage.
As I noted on the web page, I didn't have much choice there because the selection of small-signal MOSFETs is very limited; all the more powerful ones I could find had high input capacitance. Also, this is a very low power amplifier (25W into 4R), yet still actually a lot more powerful than necessary due to high speaker sensitivity.Lumba Ogir said:
I see what you mean there, but because of the way the cascode bias voltage is taken from the LTP, they need high gate input impedance, otherwise they end up affecting their own input current. Indeed, my initial tests indicated that MOSFETs worked slightly better than BJTs. The distortion of the amp as a whole came out pretty low for a folded cascode, so I'm happy with the choice.Lumba Ogir said:
Hi
I use a mosfet for the VAS, but not as the amplifying devices. For those, I use THAT arrays. The drains of the cascode mosfets are the VAS output. However, the gates must have a steady fixed bias, with a relatively low Z so that the amplifying devices are driving the input capacitance via its source instead of the input stage being stuck with that duty. This is why Mr. Evil's sub amp comment refers to needing a high impedance mosfet. I used a 7002 N-type and 8002 P-type, pretty general, quite cheap. They even come as duels in a SOT-563 now!
...
The output Z of the VAS is a bit high, ~30K, but is only required to drive the bases of the first pre-drivers, whose dynamic drive current is all but negligent. The limited transconductance of these mosfets is not a problem because they're not required to work anywhere near their maximum conductance….or Pd max of a SOT-563 for that matter. BTW, >500KHz is not out of range for the circuit. When the VAS might have to 'trans-impede' a couple uA, you can use small signal stuff and build a fast and accurate VAS. Some people don't like such high Z VAS because they have to use an extra gain stage in their gnf loop, but if you don't use gnf, well .........that's another story.
I use a mosfet for the VAS, but not as the amplifying devices. For those, I use THAT arrays. The drains of the cascode mosfets are the VAS output. However, the gates must have a steady fixed bias, with a relatively low Z so that the amplifying devices are driving the input capacitance via its source instead of the input stage being stuck with that duty. This is why Mr. Evil's sub amp comment refers to needing a high impedance mosfet. I used a 7002 N-type and 8002 P-type, pretty general, quite cheap. They even come as duels in a SOT-563 now!
... The output Z of the VAS is a bit high, ~30K, but is only required to drive the bases of the first pre-drivers, whose dynamic drive current is all but negligent. The limited transconductance of these mosfets is not a problem because they're not required to work anywhere near their maximum conductance….or Pd max of a SOT-563 for that matter. BTW, >500KHz is not out of range for the circuit. When the VAS might have to 'trans-impede' a couple uA, you can use small signal stuff and build a fast and accurate VAS. Some people don't like such high Z VAS because they have to use an extra gain stage in their gnf loop, but if you don't use gnf, well .........that's another story.
Lumba Ogir said:
Perhaps you should be a little more specific and avoid generalizations? The above is your own opinion/preference.
BTW, yes, MOSFETs should be avoided in VAS stages, but your explanation as of why is flat wrong. You don't seem to be prepared to listen to others, so I'll skip the full explanation for the moment. It's enough for you to know that it's not based on listening tests or any other subjective BS and that CBS240 is essentially correct.
What sort of designs do you like then? Personally, I am pleased with it. It was a combination of several things that I hadn't used in a power amp before: Complementary JFET input stage (not practical until I found a source of complementary matched pairs - a source which has unfortunately dried up again); Adding gain to current mirrors to allow their use for loading the LTP in a symmetrical folded cascode; and small-signal MOSFETs. I also thought it rather neat the way it ended up requiring no external current or voltage sources for biasing.Lumba Ogir said:
Sure, a folded cascode is not the best choice for an audio amp - I could have achieved better measured performance from only common cheap devices if I had used a standard three-stage topology, but I would have learnt nothing from that.
They certainly aren't noise-free (I even chose green LEDs for aesthetic purposes rather than the quieter red ones), but then again they aren't used in a position where noise matters. Yes, I used ceramic capacitors for supply bypassing, but I still don't see why that's so bad. The performance of the amp is not lacking in any way.Lumba Ogir said:
CBS240,
the input impedance should be low. High gain, small reverse transfer capacitance and small conduction losses are further requirements, metaphorically, a medium where voltage swings can move uninhibitedly. The resistive nature of FETs does not promote that. They can impossibly perform both conversions equally accurately, can they? Optimized BJTs are preferable.
Mr Evil,
syn08,
diodes and ceramic materials bring about unpleasant distortion, I was not the first to make the observation. Measurement data do not tell the sonic impact of material nonlinearity, without listening test you will never know. For a minimum of distortion, dielectric losses and coloration in capacitors, cables and sockets I recommend Teflon, thank me afterwards.
the input impedance should be low. High gain, small reverse transfer capacitance and small conduction losses are further requirements, metaphorically, a medium where voltage swings can move uninhibitedly. The resistive nature of FETs does not promote that. They can impossibly perform both conversions equally accurately, can they? Optimized BJTs are preferable.
Mr Evil,
I would not lament excessively.The complementary LTP is far from the best sounding one. The fewer devices carrying the signal waveform the better. Consider a (British-style) single-ended input.
syn08,
diodes and ceramic materials bring about unpleasant distortion, I was not the first to make the observation. Measurement data do not tell the sonic impact of material nonlinearity, without listening test you will never know. For a minimum of distortion, dielectric losses and coloration in capacitors, cables and sockets I recommend Teflon, thank me afterwards.
My explanation is deficient, but not wrong.
I agree, this makes sense for the VAS amplifying transistors. If you choose to cascode the VAS, a mosfet could be used as the cascode device if the conditions are right. In my circuit, the VAS output is clamped before the mosfets reach saturation so Crss affects very little. Crss in these mosfets might be 10pf, but is driven by the amplifying transistor not the input stage, or common gate. Separate PS means I can sacrifice a few volts, no big deal. Easier to get rid of them than create them at such relative small currents. I tried both BJT and small signal mosfets, and the difference was small, but with the mosfets, I was able to squeeze out an extra couple hundred KHz or so bandwidth.
Lacking minority carriers is one reason they can be faster. The downside is finding a small signal mosfet with higher Vds than 60V.
And it would be cool to try depletion mosfets, but the p-type seems to still be made of ‘unobtainium’. Kinda limited there. 
IMO, it's not how many devices but how far and where the Q-point varies in each device. This is one advantage of cascode, it allows you to better steer the Q-point of your amplifying transistor to a more linear transition, for a few reason I'm not going into. There is no rule that says you cannot direct couple complete amplifier stages together. Sort of a network like thingy.
It all depends on your personal POV and weather or not this is a home project or you have some A-hole boss breakin' your balls over it. My opinion on cap type, polystyrene, polypropylene, and tiny SMD COG type pf caps for compensation. But you low impedance signal guys don't like them because they do nothing for you.
The complementary JFET LTP has an appealing simplicity about it. "Best sounding" is a non-issue here because, as with an decent low-distortion amplifier, the differences are way below perception. It is effectively 100% transparent.Lumba Ogir said:
I have never seen any double-blind test that showed that anyone could detect different capacitor types except under the most extreme of conditions where the frequency response was affected. And no, I've never heard any differences myself either.Lumba Ogir said:
Anyway, this has drifted off-topic now.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.