Alternate PWM modulation schemes

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
With H-bridges there are 3 common ways modulate the bridge these are:
High side Recirculation
Low side Recirculation
and No recirculation ( one side or other of the bridge is always active)

I've noticed that most of the ClassD designs being discussed avoid current recirculation modes by switching equally between the two conduction states to achieve zero output.
My spice simulations show significant power and distortion advantages in the use of a waveform similar to the double sided Natural sampling waveform proposed many years ago by Karsten Nielsen, Yet I don't see this as the dominant form for ClassD modulation.

Clearly there needs to be some form of correction for the minimum pulse width problem, but this has several obvious fixes.

I see three main advantages:
Lower EM emmissions
Lower Average Power Losses
Lower open Loop distortion

What am I missing?

ClassDunce
 
I believe what he means

Is that the amp has three states, instead of 2: Inductor to V+, Inductor to V-, or inductor to GND. At each rising clock edge, a decision is made: The output needs either more or less current. Either of the output devices is switched on for a period, then switched off. The device is set up such that no matter where you are on the signal, there is always an off time. In this idle period (off time), the inductor is connected to ground.

This makes for a lower noise environment when the PWM comparator is making its decision for the next cycle.

Furthermore, the on time can be adjusted based on the output amplitude. For low level signals, the on time can be minimal. For zero output, the device pulses high for the minimum on time, waits a full clock cycle, the pulses low for the minimum on time. The RMS energy is much less than the traditional 50% square wave.
 
"Three state" and "four state" H bridges

There is a popular ways of achieving this. The approach used by Apogee (DDX) and TI use the two sides of the H bridge driven antiphase in one direction for a "positive" pulse, antiphase the other direction for a "negative" pulse, and both sides driven low to create a low impedance "zero" state. The modulating scheme is such that at low/zero output voltage, one gets short pulses of alternating polarity. At higher levels, the output switches between "zero" and the desired rail. This significantly lowers the total RMS current flowing and is what allows the TI low power amps to get away with no (or minimal) output filtering without destroying the loudspeaker load.

A couple problems with this: One has some intersting problems getting a smooth class AB transition through this region. The best solution I've seen was developed by Norman Crowhurst. His 1964 design (no typo, 40 years ago) amplitude modulted these narrow pulses to get finer control in the crossover region, where there is little issue with efficiency. (See his patent, http://www.acutechnology.com/ClassD/pat.pdf/US03336538__.pdf)

Another problem is that this scheme looses one of the nice benefits of the H Bridge architecture: The output impedance in an H Bridge is the sum of the On-resistance of one upper and one lower FET in the bridge. Its easy to keep these matched, even if the upper and lower FETs or their drive are different. When you have this "zero state" with the equivalent resistance of the two lower FETs, you tend to have more problem with time/signal dependent output impedance, which gives distortion (third order in this case).

This also kills the tendency for "self commutation" where the reaction of the output inductor to the turn-off of one device is to automatically swing to the opposite rail, independent of the delay before the opposing FET turns on. This makes the signal-dependent delays symmetrical for each transition. As the output current exceeds the nominal "circulating current", then the inductor current will be such that after one device turns off, the voltage stays put untill the opposite switch turns on. This will happen on one but not both edges, giving a pulsewidth error rather than just a delay. Despite potential power loss, this recirculation phenomenon gives the best output stage distortion when the circulating current is high.

(Three state switching output stages go back at least to '62, when this technique was applied to switching amplifiers used for "inverters", DC->AC converters that replaced motor/generator sets.)
 
Sorry everyone for not following this forum more closely, truth is that I had a tapeout deadline that jumped up and consumed all my time.

Karl,
Thanks for your detailed reply, the TI solution this is similar to what i am talking about. You are right to mention the transition region where there is certainly an added distortion, however there are several ways around this.

1) always have a minimum pulse width and add to this the delta width resulting from the actual pwm modulator. this is simple to achieve with a negative triggered one shot (falling edge of actual PWM triggers min width pulse) than merge the two into a single pulse for the Hbridge and trigger a second one shot for the opposite side.

2) I see no good reason to stop doing this until you reach say 90% modulation, at this point all the other distortion sources are also increasing. So the AB transition happens at high volume, which results in a decrease in efficiency at mid power levels.

3)The RF interference from this signal looks like an AM signal with suppressed carrier. This results in much better "whitening" of the EMI just due to the polyphonic nature of music. With standard ClassD there is always a big carrier component to the EMI.

4) I accept your point about an added distortion component that is due to the impedance of the two low OR two high switches not exactly matching the High + Low impedance. I'll need to run a few more simulations but I think I now understand why TI is switching between alternate high side and low side recirculation? hmmm (Silly me i thought they just enjoyed debugging CMOS latchup problems)

5) Self Commutation effect, I had not seen this effect because i was modeling with ideal switches, however with real Fets there is certainly a distortion component associated with this effect. Interestingly the distortion is not present at very low signal levels because both the positive and negative side errors match (assuming a min pulse is always added / subtracted from the actual PWM) But as you increase the signal level you transition from "discontinuous conduction to continuous conduction" using the SMPS terms, right at this transition point you have an added distortion, you can see it as a step change in the gain transfer characteristic. Bummer ....I don’t have a solution for this effect

I remember hearing that TI had to do a lot of work to get their Hbridges to switch in under 5nsec under all cases, I wonder if this was their main problem. Unfortunately for Audio applications this transition occurs at low to moderate power levels so it will peak in the standard listening range.

ClassDunce
 
Re: "Three state" and "four state" H bridges

kral said:
....A couple problems with this: One has some intersting problems getting a smooth class AB transition through this region...


Karl:

You mean the nonlinear width / recovered mean load voltage transfer caused by flywheel diode conduction?

If that were the case, I tested an idea a couple of years back in a 60W single ended prototype. It was to place in paralell with the output (before LP filter) a low Q HP filter (capacitor in series with resistor and inductor in paralell).

Both in simulations and actual operation, output device disipation was considerably reduced and linearity improved not having the diodes conducting.

Efficiency deteriorates by not pumping back HF energy (carrier and harmonics) though, simply diverting it partially to the resistor.

Rodolfo
 
Karsten Nielsen's Work

His theories are implemented in the ICE Power modules. B&O were so impressed, he was given some sort of ownership poition in the company. I know I'm impressed with the current production 1000A and 500A modules. But then, I'm an OEM and I can actually buy them. I've used several thousand the last three years.

For the rest of you, I have tested the UcD180 from Hypex and I find them quite good as well. And they are easier to connect as they don't need external ±12V supplies.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.