IcePower slew rate

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
The concept of slew rate does not make sense for class D amplifiers, or even for modern linear audio amplifiers at all. It's mostly operational-amplifier related, where it still makes a lot of sense, particularly for non-audio applications, where waveforms can have sharp edges.

Audio people started to consider slew rates many years ago in the context of linear amplifiers that could not produce full output up to 20Khz, (or whatever upper frequency was required in each application, not only audio!). This was because some internal stage would clip before delivering enough current to charge the input capacitance of the next stage fast enough. This results in "triangulation" of the output waveform (when maximum slope is reached).

Nothing of this exists on class D, there are no linear power amplification stages and no triangulation is possible (at least not unless the designer did something really wrong, like using a too big output capacitance with a too low current limit, so unlikely).

Anyway, modern audio amplifiers, both class D and linear, can produce full output well above 20Khz, so they cannot distort the signal in the way the old "slow" ones could do, and by "slow" I mean stuff from LM741 age, 1970s, germanium, tubes with high capacitances (yes tubes!), etc.., so way way outdated!
 
  • Like
Reactions: 1 user
Eva, I fully agree. If an amp can produce full power at 20 kHz with low distortion (and full stability at actual speaker loads), that's already in excess of anything that will ever appear in real music signals.

Reminds me a little bit on the damping factor discussion, where totally meaningless numbers are claimed by not including the voice coil resistance.

Kurt
 
I agree with Eva regarding 'slew-rate' in class-D amplifiers. Just consider: the output of the switching transistors (at the node before the output LP filter) is switching from rail-to-rail at a slew rate that is orders of magnitude higher than the rate-of-change of any audio output signal. This switching signal is then low-pass filtered by a passive filter that imposes a 2nd-order response, but cannot cause any slew-rate limiting in itself.

I don't know the details of the self-oscillating modulator used in IcePower amplifiers, and it is conceivable that the modulator stage (if not designed properly) could cause a slew-rate limit, but even if it did, it should be appreciably higher than any audio signal will require.
 
In this case I agree with Gyula. I think if slew rate is limited, then it's slew-rate limiting. And slew rate is limited indeed. Limitation comes from the supply voltage, but it exist. The saturated element is PW modulator (at 0% or 100%).

In some cases the power bandwidth can be significantly lower then small-signal bandwidth (because of the efficient feedback), so I think SR is a relevant parameter in ClassD also. But it's soooo low (compared to linear amps), that nobody dares to specify it. :)
 
Pafi said:
In this case I agree with Gyula. I think if slew rate is limited, then it's slew-rate limiting. And slew rate is limited indeed. Limitation comes from the supply voltage, but it exist. The saturated element is PW modulator (at 0% or 100%)...
The two are different! I've attached a sketch showing the difference. With a pulse input (top), a low-pass filter results in a curved output (middle), whereas a slew rate limit results in the edges being limited to a fixed slope (bottom).
 

Attachments

  • limited.png
    limited.png
    5.5 KB · Views: 567
At linear amps the SR limit not neccessarily manifested in triangular waveform neither. This is just an idealisation. In reality maximal SR doesn't have to be constant! If there is a limit, then it's done. It can be a function of load impedance, supply voltage, instanteous output voltage, previous states, etc..., but if increasing input doesn't make it faster, then it's slew rate limit. At least I call it this way. (And semiconductor companies too, because otherwise they couldn't specify some of their OPAs.)
 
Pafi said:
At linear amps the SR limit not neccessarily manifested in triangular waveform neither. This is just an idealisation. In reality maximal SR doesn't have to be constant! If there is a limit, then it's done. It can be a function of load impedance, supply voltage, instanteous output voltage, previous states, etc..., but if increasing input doesn't make it faster, then it's slew rate limit. At least I call it this way. (And semiconductor companies too, because otherwise they couldn't specify some of their OPAs.)
That the slew rate is not constant does not change anything - the fact remains that both the cause and the effect are different. If manufacturers are using the term differently then they are being misleading.

Eva mentioned sine waves, so perhaps those will illustrate better, as in the attachment. In a band limited system, the sine wave will at worst be smaller in amplitude and phase shifted. On the other hand, if subject to a slew rate limit then it will be distorted (not necessarily perfectly triangular, as you say, but the output will contain new frequencies = distortion).
 

Attachments

  • limited_sine.png
    limited_sine.png
    6.1 KB · Views: 427
On the other hand, if subject to a slew rate limit then it will be distorted (not necessarily perfectly triangular, as you say, but the output will contain new frequencies = distortion).

I agree! And this can happened to ClassD also. (Only it's hided by the filter, but if you look at the pre-filter stage with FFT, then you can see the distortion. You can measure the distortion after the filter too, just it's too damped to be seen by eyes on the scope.)

the fact remains that both the cause and the effect are different.

Different from what? From a specific case of the SR limit? Yes.

Generally: there is a saturating stage, and after an integrating network in both cases. In this case the integrating network is 2'nd order, and not perfect, these are the only difference from the usual.
 
You would have to fed the amplifier with an input voltage outside the "passband" higher than the voltage required to clip the amplifier on the passband. Otherwise you just get an undistorted lower amplitude.

Not true. For example build an UcD with an output LPF cutoff freq of 15 kHz, but set the first corner freq of the feedback to 40 kHz! The amp will have a small-signal bandwidth of 40 kHz, but it will hardly saturate with 0 dB 40 kHz input signal.

Anyway: if output current is limited to Imax, then slew rate is limited to Imax/Cout.
 
I think that you can't really do that with UcD. At least not in a practical way (active filtering of the carrier residual before feeding it to the comparator is very risky due to the low delay and the high bandwidth and low distortion required ;) )

I've used both 1st order and 2nd order and the feedback corner frequency is inherently very close to filter resonance in both cases ;) In fact, it's difficult to get it close enough (because it's below).

Remember the phase lead network that kicks in at filter resonance and above, it cuts gain too...
 
At least not in a practical way (active filtering of the carrier residual before feeding it to the comparator is very risky due to the low delay and the high bandwidth and low distortion required )

Check this circuit!

Actually this is a very practical thing, and this way you get higher open-loop gain. But it has a lower power-bandwidth than the small signal one.

Remember the phase lead network that kicks in at filter resonance and above, it cuts gain too...

Do you mean R2-C2-R4? This is what I was talking about. You can set it much above resonance, this way it doesn't cut significantly.
 

Attachments

  • sr limited ucd.gif
    sr limited ucd.gif
    5.2 KB · Views: 377
This seems to be intended to self oscillate at 500Khz, way way above filter resonance. With such an arrangement you would get 25mV p-p of carrier residual at the comparator so it's not very practical or reliable (or low noise).

Also the filter is not properly designed for 4 ohm operation, it's 4 dB down at 20Khz.

The relationship between filter resonance frequency and the frequency at which the phase-lead zero is set also has a dramatic impact on the amount of self-oscillating frequency drop near the rails. The highest you set it, the more drop...

The integrator(s) or pole(s) are missing. They are essential to provide HF gain roll-off to match the response of the filter.

Indeed, there are optimum ratios and topologies for all that (Bruno got quite close) and with these and a load impedance within rated it does not triangulate, even if frequencies above the passband are fed.
 
  • Like
Reactions: 1 user
This seems to be intended to self oscillate at 500Khz, way way above filter resonance.

Slightly below 500 kHz, but never mind! Do you remember what is the discussion about? Not if high osc. freq. is good or not, but if there can be a ClassD amp with less power bandwidth then small-signal bandwidth. And why do you hurt that 500 kHz? Just think of the 'MHz amp'!

With such an arrangement you would get 25mV p-p of carrier residual at the comparator so it's not very practical or reliable (or low noise).

I don't really agree, but this is out of the original question. Who cares the osc. freq within this argue? You can increase R4, and you get lower freq, but every other things stays the same! And the question is not about specific values and layout difficulties, but about principles.

Also the filter is not properly designed for 4 ohm operation, it's 4 dB down at 20Khz.

Good morning! This is what I am talking about! This is exactly why a ClassD amp needs further specifications over 'bandwidth'. Now you can see the filter, because I showed the schematic with values. But if this was a real amp within a black box, you couldn't see the values, and if you saw only the bandwidth spec., you weren't able to decide if it is a good amp or not! (However it would be perfect for bass or mid-range.)

But if you are irritated by this filter even in theory, then you can change everything for higher freq, it doesn't matter. I've built a similar amp some years ago, with 30 kHz filter, and >50 kHz small-signal bandwidth. It works perfectly.

The integrator(s) or pole(s) are missing.

Yes, and there is no overload protection. Who cares? This is a simulation of the principle. Place an integrator, if you want, it won't change any important thing! The basic UcD doesn't have an integrator either.

They are essential to provide HF gain roll-off to match the response of the filter.

Que? Match to what?

Indeed, there are optimum ratios and topologies for all that (Bruno got quite close) and with these and a load impedance within rated it does not triangulate, even if frequencies above the passband are fed.

Again: the discussion is not about 'optimal' ratios (whatever it means), but about SR limits of ClassD generally. Specifications are not only for optimal amps, but every amps!
 
All is related in this topology. Filter resonant frequency, phase-lead compensation, gain reduction near filter resonance and switching frequency.

A lower and more close to optimum Fsw, with proper phase margin for low Fsw drop (and reliability with no load), results in substantial gain reduction at 20Khz (due to the phase-lead network required) and the modulator no longer trying to get an output voltage level that the filter can't provide on 4 ohm. Try to simulate the values and pole topology shown in the classic UcD paper (if you haven't already done it a hundred times...)

The maximum output that the filter can provide rolls-off at 12dB oct, so the integrators should provide the same gain roll-off.

The L and C have to be sized to still get some resonance boost with the minimum rated load. No need to change Fres.

Of course, I have got almost load-independent flat-to-20Khz response from clocked PWM voltage / average-current loops with filters resonating as low as 3Khz, but such a system does triangulate (and does not make much sense as a full-range amplifier, I use it for sinewave inverters).
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.