Designing a line-level signal stage with high enough PSRR isn't too difficult - just make it classA and feed it with a low enough noise supply. Current invariant classA means no self-generated supply noise so only the reg's noise matters. This gets a lot tougher with amps unless they're classA, so I'd say the bottleneck is most likely to be the amp. Using opamps though potentially reverses that because they're normally classAB and are often given heavy-ish resistive loads to get good SNR figures, along with power supplies designed by the numbers.
Last edited:
As an introduction to those not well versed in the art, and as an aid to get everyone to state their assumptions I like this page LM3886 Chip Power Amplifier Power Supply Design. . Although written for chip amps all the maths is entirely transferable and in a far more digestible style than I can manage.
At least with some requirements you can define if the amplifier is fit for purpose. "can't hear a problem on unspecified speakers, with unspecified music at unspecified SPLs and lost my notes anyway" doesn't help us reach a common ground.
At least with some requirements you can define if the amplifier is fit for purpose. "can't hear a problem on unspecified speakers, with unspecified music at unspecified SPLs and lost my notes anyway" doesn't help us reach a common ground.
Ed
The tweeter drops to a Z of 2.2 Ohm at 18.6kHz. It is 1.76Ohm XR and 1.43 Ohm XL, so only slightly reactive. But what is worrying and I haven’t measured is if the reactance goes sharply up above 20kHz.
From my notes, the measurements were performed with all the adjustment resistors shorted. The ohmic component then is all due to T27.
Rdc was 5.9 Ohm for T27, well within it’s specs
https://www.madisoundspeakerstore.com/pdf/kef/Kef Driver Spec Sheets.pdf
The dummy loads were all 8.8 Ohm
Attached are all measurements (REW V5.13) that can show what affects which.
For the record, to feed the current discussion on amplifiers, these speakers at the home of their owner, were driven quite well with a Kenwood KA-5700 but with a Marantz PM-47, sound was anaemic. I have opened both (same quality of construction), I have looked at their specs (same output specifications) and their schematics (power output stages topology, transformer size/volts, smoothing caps C, power rails V are the same).
Only the Marantz is at least 15 years younger
George
Attachments
The point of intelligent engineering is to look at the overall system to get a targeted result- PSRR in isolation is meaningless.
Thanks George,
The HF driver is indeed more of a load than the resistor. I would not expect a low efficiency soft dome tweeter to show the back emf drop. In the more common resonance the model (which I really don't like) is of a parallel RLC network. The other version of course is the series version. To get this behavior I believe requires a very stiff piston.
When I have time at my office I'll look for an example of a drop in impedance below DC resistance of a driver. Note in most of my work we don't use passive crossovers.
The HF driver is indeed more of a load than the resistor. I would not expect a low efficiency soft dome tweeter to show the back emf drop. In the more common resonance the model (which I really don't like) is of a parallel RLC network. The other version of course is the series version. To get this behavior I believe requires a very stiff piston.
When I have time at my office I'll look for an example of a drop in impedance below DC resistance of a driver. Note in most of my work we don't use passive crossovers.
Last edited:
I agree . . . PSRR in modern opamps is generally beyond reproach. Discrete designs may be much less of course - but then you can always engineer a good supply to offset that weakness IMV. Only think to be careful with is decoupling on opamps supplies - Kendall has that one covered.
If we say that THD or any signal error is the "sound" of the system, then we can look at the individual sources of this error. If we say that PSU noise gives one sound and amplifier error gives another sound, then we can say the louder one will mask the other. Therefore, it makes sense to choose the error you prefer and let it mask the other errors.
Of course if we accept the idea of masking then we must consider that the error may be masked by the signal.
Of course if we accept the idea of masking then we must consider that the error may be masked by the signal.
For PSRR you need to include frequency in the spec. Let me have a go at this:-
Practical VFA Power Amp - >70dB up to 1kHz, dropping off above that at 20 dB/decade. Add a Cap multiplier for better performance.
For CFA, drop the figures by 20 dB
For opamps - modern exemplars are doing 140dB at LF, with >100dB at a few kHz and 70 dB at 100 kHz - I'd classify their performance in the 'problem solved' category wrt audio.
Practical VFA Power Amp - >70dB up to 1kHz, dropping off above that at 20 dB/decade. Add a Cap multiplier for better performance.
For CFA, drop the figures by 20 dB
For opamps - modern exemplars are doing 140dB at LF, with >100dB at a few kHz and 70 dB at 100 kHz - I'd classify their performance in the 'problem solved' category wrt audio.
You can't answer that in isolation. As mentioned before, the crap that ends up on the output of your amp is the product (quite literally) of PSRR and psu noise. The two are interchangeable and a compromise has to be selected.
The only meaningful requirement is the S/N ratio from the psu point of view. if you want, say, 100dB, it can be reached with really high PSRR paired with a so-so psu, or a really good psu paired with so-so PSRR, or something in between. It's up to the designer where he spends his money.
Jan
For virtually any design I think it's wise to do simulation studies to get a prediction of behavior, including power supply rejection with frequency. There are opportunities, usually, to better the performance via changes in the topology, as opposed to brute-force reduction of power supply hum and noise. Likewise, in terms of sagging rails, their effects can be studied in simulation too.
I don't consider an audio design as ready for prototyping until one has looked at a variety of such aspects. One frequently overlooked: the power cycling transients and whether base-emitter junctions are being reverse-biased to breakdown or DMOS gate-source breakdowns are being exceeded.
Considering the audible importance of clipping behavior, it seems strange that so few products provide a visual indication of clipping. It is as if no one wants to know.
I don't consider an audio design as ready for prototyping until one has looked at a variety of such aspects. One frequently overlooked: the power cycling transients and whether base-emitter junctions are being reverse-biased to breakdown or DMOS gate-source breakdowns are being exceeded.
Considering the audible importance of clipping behavior, it seems strange that so few products provide a visual indication of clipping. It is as if no one wants to know.
How strange that you should mention that. This is something I've had to look into on one of my designs whereby the B-E junction of the input stage is reversed biased for 'some considerable time' during power on. Its not destructive at the currents involved but it can I believe seriously degrade the noise performance of the device. Better to get it correct in the initial design than for it to surface as a problem at some later date.
Yes, it is a serious degradation mechanism. Motchenbatcher cautions about testing low noise bipolars with an ohmmeter in his first book. He also mentions that the damage done can be annealed out by high temperature baking, although this is scarcely practical!
Nope, just plain engineering sense. Not common in fashion audio, I'll grant you.
In view of the above, and as per your experience, Andrew, how do you feel about split supplies for the voltage amd and the current stages? Using say a "virtual battery", which is basically a capacitance multiplier governed by a zener diode setting of the voltage?
Actually, "virtual battery" is not a technical, but a trade name used by Matsushita/Technics. They contend that if the power transistor used is suffciently more powerful than the worst case current demand and is followed by a large value cap (say, 1,000 uF or more), the regulation is such that it appears to be more like a battery than a regulator circuit (their words, not mine).
I've heard that too (ohmmeter testing) although I can't recall where I read it, and... you've just reminded me of something
the AVO 8 comes to mind with its 15 volt battery on high ohms range. I remember at college and how we had to test and sort transistors into functional/non-functional and NPN/PNP. Back then they were all metal BC107/BFY50 packages or TIP41/42 with numbers rubbed off and I had learnt the trick of using an AVO on high ohms to see if the device was NPN or PNP simply by observing reverse leakage from C-E when reverse biased. Get the polarity correct from that and one more test of a wet finger dabbed across C-B saw the needle swing hard over to zero if the device was good.
My classmates were amazed I could sort them all in seconds while they all followed the standard 'diode checks' across every junction.
Happy days
A BJT C-multiplier output impedance looks like a diode. No matter how powerful the BJT, you get basically the same current-dependent output impedance. Unless of course the transistor is severely underrated and then it's more constant. If by more like a battery they mean significant output impedance then they are correct.
power supply rejection of simple emitter follower stage
I did a bit of sim with a voltage-source-driven complementary emitter follower output stage. I used the ancient 2N3055 and MJ2955 (as used in the NAD 3020) with 200 milliohms in each emitter and voltages across the bases to establish a fairly rich 200mA quiescent bias, and run from +/- 30V rails, and with an 8 ohm resistive load. The power supply rejection was about 80dB out to a few hundred kHz, with the negative rail having somewhat better rejection than the positive.
For a 1V peak input at 1kHz, the distortion out to 15th is predicted to be 108ppm. Due to less than perfect complementarity, the dominant component is second harmonic (at about -79dB, with third at about -92dB, and higher harmonics much lower).
I added 2 ohms in series with each supply. For 1V p-p inputs this raised the 1kHz distortion by a very small amount (to 110ppm). At high signal voltages the 2 ohm Rs cause the collectors to crash into the emitters as it were, and of course then the distortion rises precipitously, while the zero impedance supplies version keeps on going for a while.
Again, note the assumptions: no thermal distortions, and voltage source drive. But it does confirm what one would expect, that due to the relatively high impedance at the collectors, the sensitivity to power supplies' noise and impedance below overload is pretty small. The voltage source drive is unrealistic and conceals other mechanisms like base width modulation that will degrade performance for more realistic base drive. But we can come close.
I did a bit of sim with a voltage-source-driven complementary emitter follower output stage. I used the ancient 2N3055 and MJ2955 (as used in the NAD 3020) with 200 milliohms in each emitter and voltages across the bases to establish a fairly rich 200mA quiescent bias, and run from +/- 30V rails, and with an 8 ohm resistive load. The power supply rejection was about 80dB out to a few hundred kHz, with the negative rail having somewhat better rejection than the positive.
For a 1V peak input at 1kHz, the distortion out to 15th is predicted to be 108ppm. Due to less than perfect complementarity, the dominant component is second harmonic (at about -79dB, with third at about -92dB, and higher harmonics much lower).
I added 2 ohms in series with each supply. For 1V p-p inputs this raised the 1kHz distortion by a very small amount (to 110ppm). At high signal voltages the 2 ohm Rs cause the collectors to crash into the emitters as it were, and of course then the distortion rises precipitously, while the zero impedance supplies version keeps on going for a while.
Again, note the assumptions: no thermal distortions, and voltage source drive. But it does confirm what one would expect, that due to the relatively high impedance at the collectors, the sensitivity to power supplies' noise and impedance below overload is pretty small. The voltage source drive is unrealistic and conceals other mechanisms like base width modulation that will degrade performance for more realistic base drive. But we can come close.
- Status
- Not open for further replies.