an audio amplifier is, fundamentally, a voltage controlled current source
of course the power supply is very crucial.
i for one am putting 2*15,000 μF on my 2*40 W chipamp
yeah the capacitors themselves will cost at least twice as much as the IC
Probably a good idea, although I strongly suggest using multiple smaller caps in parallel instead of one big one, for each 15kuF. And if you have several inches or more of supply and ground rail conductors, then you should still have a relatively-large capacitance much closer to each power pin.
The important question is "where?". Those 15kuF caps will have large diameters and lead-spacings. So even if you tried to get them as close as possible to the chip, it could not be close-enough to satisfy all of the transient-response requirements.
You will need, for example, at least 220 uF within one inch of each power pin (and more, farther away). But because the inductance would then probably still be too high, you would need to use three 68 uF caps in parallel instead of one 220 uF cap. And the three 68 uF caps would need to have three separate conductors all the way to the power pin.
Of course that is all just somewhere in the ballpark, on my part, from a chipamp example I worked out. It will depend on your choice of acceptable maximum supply rail voltage variation (and your chipamp's max slew rate and rail voltage and load impedance).
But the equations for figuring all of that out, "exactly", are not hard to apply and almost all you need is in the posts I linked to, earlier (although as I said I did not yet finish compiling the whole methodology, even for just diyaudio purposes).
As I mentiond earlier, with the frequencies and layout geometries that are used in diy audio, it looks like it is very easy to not do it quite well-enough, in which case your transient response would not be as accurate as it could have been.
For the "simplified version" of the calculations (which is basically all I have, so far), all you need to know are: 1) the maximum current change, Δi, that could happen, 2) the time interval, Δt, during which it might need to occur, and 3) your choice of acceptable rail-voltage disturbance, Δv. Then you can calculate the minimum decoupling capacitance required:
C ≥ ( Δi ∙ Δt ) / Δv
which is a direct approximation of i = C dv/dt, by the way. So far, this version is meant to be applied only over short-enough time intervals.
The maximum tolerable inductance for the decoupling cap configuration will determine the maximum trace or wire length that can be used:
L ≤ ( Δv ∙ Δt ) / Δi
When looking at the self-inductance of the pcb traces or wires, a common estimate is roughly 15 nH per inch. You can also estimate the inductance for the capacitor, using 15 nH per inch of lead-spacing.
With the Δv and Δi, you can also think in terms of a target impedance, at the device power pins, for meeting the transient response requirements:
Ztarget ≤ Δv / Δi
which must be valid up to a frequency of at least
f = 1 / ( π ∙ trise)
where trise will probably often be the same as Δt. That equation gives the highest Fourier frequency component needed in order to construct a pulse-edge with a rise-time of trise.
I am doing all of this from memory so you should really read the other links I have posted, where everything is also tied together much better, and additional parameters and equations are considered.
But, as a disclaimer, even for those posts I never did find the time to go back and really finish the job, so there are a few "loose ends" that remain. ESR is one that I started to account for but didn't get back to.
Also, for example, how big can Δt be for the L and C equations above to still be good-enough approximations? I don't know yet but at some point, as Δt is increased, i.e. at a low-enough frequency, the allowable inductance will have increased enough that it is more than the inductance due to the length of the supply rails, and then the main reservoir caps will be able to supply the current accurately-enough, and without causing too much Δv on the supply rail.
EDIT: -------------------------
Actually, I just realized that's a key point and we need to turn that question around!! How large do we HAVE to take Δt, for the decoupling cap calculation? After all, increasing it makes the required capacitance larger, which makes the physical geometry/layout problems more difficult, in general.
It looks like the answer is to estimate or calculate (or [gasp!] measure) the total self-inductance of the supply and ground rails (round trip) and then use the (or "an") inductance equation to calculate the minimum Δt for that inductance, which would be the shortest time in which the worst-case current change (the Δi we specified) could be provided through the power rail instead of from the decoupling cap, without causing a rail voltage disturbance of more than our desired Δv. (Woo hoo! One more "loose end" tied up!)
So that also means that longer supply rails from a psu make it necessary to use larger decoupling caps, since the longer the rails are, the lower the frequencies are that the decoupling caps must also be able to supply.
In reality, or in a practical design situation, we would probably often do that type of calculation "backwards", and determine the frequencies we needed the main reservoir caps to be able to handle and calculate the rail length from that, instead of the other way around.
So now we know that there is a "crossover" type of relationship, between the reservoir caps and the decoupling caps.
------------------------- /EDIT
Well, anyway, at least the equations above go in the right direction, as Δt increases. i.e. As Δt grows, you need more capacitance, but also more inductance is able to be tolerated (i.e. the larger caps can be farther away for the lower frequencies).
Maybe in most practical layout-design situations, if we do some preliminary calculations and start to see that it might get difficult to fit enough decoupling caps close-enough to a device's power pins, then we might decide to first determine what IS physically possible (or convenient) and then calculate backwards to see what kind of hit we might have to take, in terms of transient response performance.
Sorry to have blathered-on for so long about all of that.
Regards,
Tom
Last edited:
Whatever the ripple factor that seems acceptable
it is important to design a power amp PSU such that :
- Supply voltage is no more than 0.5 time the caps nominal voltage.
- The caps temperature doesnt exceed 40°C , hence my critic of LazyCat
implementation where the caps are too close of the thermaly hot components.
it is important to design a power amp PSU such that :
- Supply voltage is no more than 0.5 time the caps nominal voltage.
- The caps temperature doesnt exceed 40°C , hence my critic of LazyCat
implementation where the caps are too close of the thermaly hot components.
I see your last point, 150 degree caps may serve better but one needs to remove any and all excessive heat, I agree.
A 60Hz signal charges the capacitor bank 60 times a second.
A 20KhZ signal drains the water tank 20,000 times per second. However it takes small bites whose size in relation to charging cycle is (60/20000). Thus (60/20000)*20000 = 60 times per second
A 20HZ signal drains the water tank 20 times per second. However it takes large bites whose size in relation to charging cycle is (60/20). Thus (60/20)*20 = 60 times per second
For us to have a buffer in the tank, the fill rate has to be larger than drain rate. The drain rate is set by the 8 Ohm speaker load. Thus we need a much lower capacitor impedance than this lets say 1/100 * 8 = 0.08Ohms
We also want to set the high pass filter such that the flow to load is not interfered with. If 20Hz is the lowest frequency that we will ever draw from the capacitor bank. We can then set our high pass filter at 1/10 * 20Hz = 2Hz
2Hz @ 8oHms = 10,000uF. However if you have four amplifiers connected to same PS then its 10,000 * 4 = 40,000uF
Thus we have not used wattage in any of our calculations
A 20KhZ signal drains the water tank 20,000 times per second. However it takes small bites whose size in relation to charging cycle is (60/20000). Thus (60/20000)*20000 = 60 times per second
A 20HZ signal drains the water tank 20 times per second. However it takes large bites whose size in relation to charging cycle is (60/20). Thus (60/20)*20 = 60 times per second
For us to have a buffer in the tank, the fill rate has to be larger than drain rate. The drain rate is set by the 8 Ohm speaker load. Thus we need a much lower capacitor impedance than this lets say 1/100 * 8 = 0.08Ohms
We also want to set the high pass filter such that the flow to load is not interfered with. If 20Hz is the lowest frequency that we will ever draw from the capacitor bank. We can then set our high pass filter at 1/10 * 20Hz = 2Hz
2Hz @ 8oHms = 10,000uF. However if you have four amplifiers connected to same PS then its 10,000 * 4 = 40,000uF
Thus we have not used wattage in any of our calculations
I agree but I'd say it differently. I think the way to think about it is that the capacitor is the device that powers the output transistor. That cap is in turn powered from either another "upstream" cap or the rectifier. (technically this is wrong, you have to look at the entire PS as an big LCR network) But if you think that way then you see why it makes sense to physically place the leads of the cap across the leads of the transistor. You greatly reduce any change of parasitic inductance in the wire. To REALLY reduce it they use surface mount caps on a PCB.
As someone here just said: "The tube guys have known this for 70 years." That is true and many times I will solder a cap lead directly to a load. In this way the power supply is "distributed" throughout the amp.
I think that we really would need to know the rail voltage, for that. If they were "big" opamps, like chipamps with 40V rails for example, more current would be required per swing and the caps would need to be larger than they would need to be if the rails were lower.
But if I'm missing something, then that would be even more interesting (to me).
I think that the "width/depth/sound-stage/imaging" is a direct result of precise amplitude timing and phase accuracy, and that very small infidelities in those will significantly degrade the soundstage imaging.
<snipped>
The rails with the battery were free of 120/100 Hz ripple, but were not unchanging. The time-varying currents of the music signal would have created a significant time-varying voltage on the rails due to Lparasitic x di/dt and Rparasitic x i. (And the amplitude of the time-varying component of the voltage would have been greater, farther away from the battery.)
But I will mostly agree and say that the power supply has at least as much effect on the character of the sound as the rest of the amp. After all, the power supply is a major component of the amp. The rest of the amp "only" controls the modulation of the power supply's currents. The audio we hear is the current that is directly piped from the power supply, through the modulated resistances of the power transistors, and to the speakers. The "small signal" path of the rest of the amp STOPS, at the bases or gates of the power transistors.
Both parts are critically important, each necessary but not sufficient by itself. But so many people don't think of the power supply as being as important as it is. So it's good to stress that idea.
Yes, try that star arrangement, by all means. For each arm of the star, put a cap close to the rectifiers and then also put enough capacitance at the other end of the arm, quite near the active device, plus some smaller values directly on the pins of each device. (I would not want to distribute rectified but unfiltered DC.)
Better yet, use smaller parallel caps at each end of each arm, instead of one per end.
You could get really ambitious and run one arm of the star to each active power output device, maybe. I'm not sure how many arms the star would actually need. Certainly you want to separate the high-current stuff from the small signal stuff, and input section from everything else. Not sure on the rest.
This might seem "way out there", but...
(For the power output devices: ) I would like to suggest that you also consider having each arm of the star to consist of two or three (or more) parallel copies of each power and ground rail pair, with a capacitance on each end of each of the parallel power/gnd rail pairs (using smaller capacitances than if you only used one set of rails, of course).
That will reduce the power supply impedance, at the load end, by a factor of however many parallel pairs of supply/gnd you put in each arm.
It really wouldn't take too much extra to implement it that way, since the caps would all be smaller, just with more of them. If it's on a PCB then each of the parallel traces could be much narrower, since you would be using several in parallel. Or, some cable with three or more twisted pairs inside might be perfect. (Or maybe even ribbon cable, as I mentioned earlier. I think I'm going to try it with 25-wire ribbon cables, and use TWELVE pairs for EACH rail, or more if the wires can't carry enough current. Maybe twelve pairs per rail per output device would be good.)
----------
Just considering real electrolytic capacitor sizes and lead-spacing and the typical geometry of a layout and the connections, etc, there must be some optimum number of smaller caps that could get a needed capacitance closest to the power/gnd pins, with the lowest minimum total inductance. I guess that number would then determine the number of pairs of conductors to use.
Last edited:
Guys, I thank you all for your participation so far, it was most enlightening. Maybe it would even be of more use to form a set of rules by combining our knowledge and group design a power supply for say a mono 100 watt rms into 8 ohm resistive load amplifier. (Equates to 28.2V rms and 3.54A rms)
I think it is a popular size amp, and lets favor our American friends having 60 Hz, while our European friends can just scale all caps up by +17% (I think)
We can also debate what the power will be into 4 ohms. We also make use of all the valid comments on the thread i.e. distributed capacitance and multiple capacitors, transformer VA rating and the like.
I think Tom, since you are the guy with a flair for calculations give us a kick start value for the main caps and the reasons for your choice and we take it from there.
What I would like at the conclusion of this thread is that we at DIY have some standard idea of what will be a good power supply for an amp that we build because that is the least discussed topic one solid state, someone builds an amp designed by a colleague and is unhappy with the product and neither understands why because they have no clue of the others power supply or what the PSU should look like.
I think it is a popular size amp, and lets favor our American friends having 60 Hz, while our European friends can just scale all caps up by +17% (I think)
We can also debate what the power will be into 4 ohms. We also make use of all the valid comments on the thread i.e. distributed capacitance and multiple capacitors, transformer VA rating and the like.
I think Tom, since you are the guy with a flair for calculations give us a kick start value for the main caps and the reasons for your choice and we take it from there.
What I would like at the conclusion of this thread is that we at DIY have some standard idea of what will be a good power supply for an amp that we build because that is the least discussed topic one solid state, someone builds an amp designed by a colleague and is unhappy with the product and neither understands why because they have no clue of the others power supply or what the PSU should look like.
Last edited:
What do you think of Rod Elliot's articles about power supply design for audio amplifiers? Personally I think they contain boatloads of info, I'd think enough to come up with a suitable supply for any size amp you want to make.
Edit: I forgot to note though that the design rules don't go much beyond common knowledge, definitely not into distributed capacitance, although even this is sort of a common practice already to decouple output stage powerrails with smaller caps locally on the board.
Edit: I forgot to note though that the design rules don't go much beyond common knowledge, definitely not into distributed capacitance, although even this is sort of a common practice already to decouple output stage powerrails with smaller caps locally on the board.
Last edited:
Okay the rule of thumb indicates 2000uF/amp thus 7200uF per rail of 47V (assuming a lateral mosfet design). The transformer should be 200VA with 0-35 & 0-35 volt secondaries. I am assuming two standard 35A block bridge rectifiers one each per secondary. snubbed with 10 nF and 10 ohm series across each diode.
The 7200uF cap per rail could probably consist of two 4700uF separated by a 0.33 ohm 5 watt resistor which will act as some low pass filter (Hugh). The second 4700uF will be constructed from 10 x 470uF caps (Tom) shunted by a 33uF and 330 nF (Harison) and could be mounted between the two output mosfets (Nico/MiiB) which could be 2SK1058 and 2SJ162.... a separate <100mA power supply producing about 51VDC per rail would supply the front end (Lazy Cat) which would use 8x Damn fast 4 amp rectifiers (Andrej) and decoupled by 10 x 220uF 63V caps (Tom) placed strategically on the PCB.
How am I doing so far?
The 7200uF cap per rail could probably consist of two 4700uF separated by a 0.33 ohm 5 watt resistor which will act as some low pass filter (Hugh). The second 4700uF will be constructed from 10 x 470uF caps (Tom) shunted by a 33uF and 330 nF (Harison) and could be mounted between the two output mosfets (Nico/MiiB) which could be 2SK1058 and 2SJ162.... a separate <100mA power supply producing about 51VDC per rail would supply the front end (Lazy Cat) which would use 8x Damn fast 4 amp rectifiers (Andrej) and decoupled by 10 x 220uF 63V caps (Tom) placed strategically on the PCB.
How am I doing so far?
Last edited:
Tom you must be itching to double the cap values for better bass punch. On the other hand Mooly would want to half the cap values to cater for "speed" and clarity. This is now open for discussion and throwing a couple of punches are allowed, we want to get to a happy medium and a rule of thumb power supply design that sports a wide range of sonic experiences.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.