why does Rod Elliot say regulating is hard on the MOSFETs

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Switchmode Power Supply For Car Audio

Two quotes from Rod's article

"For example, I use a 3:1 transformer that would give about +/-38V without regulation that is unacceptable for my LM3886 stages to be safe, so I have regulated to +/-26V. The MOSFETs will suffer more, however, so regulate the supply only if strictly necessary."

"as stated above, my version is unregulated. This will maintain maximum efficiency, and also reduces the dependence on the output filter capacitors"



Why would regulation using PWM decrease efficiency and but more thermal stress on the MOSFETs? This is not linear regulation after all. Or is it because he does not use output inductors?

Thanks
 
MOSFET duty cycle is reduced with regulation, which means the peak current is increased. Power dissipation varies linearly with duty cycle, but with the square of drain current. If power to the load is held constant, MOSFET power dissipation increases inversely as duty cycle decreases.

Output inductors even out the ripple on the output, making it easier to regulate, but don't change the story for the MOSFETs.
 
I don't think the efficiency will be significantly different, will it?

One problem is that you're sort of doing the apples to oranges thing. The power output may be greater with the unregulated 3:1 transformer than the same transformer regulated to ±26v.

There are also multiple types of regulation. You have a limiter to prevent over-voltage. You have regulation that allows an amp to double the power when the load is cut in half. There are types that work to achieve other marketing features.

I understand using regulation for some things like limiting voltage but prefer unregulated for driving speakers.
 
In the first quote, he explains that he uses regulation to make a supply for LM3886 on the fly, i.e. without changing the turns. This is presumably the first schematic, i.e. Sergio's circuit.The second quote is about a simplified circuit that he developed form Sergio's.

I agree that if all you want to do is set the secondary voltage you can do it my selecting the right number of turns. Car SMPS also tend to have excellent load regulation precisely because the transformer has a low resistance, as does the car battery. It's line regulations sucks, but at least when the engine is off or idling, the line voltage should be pretty constant (and when the car is moving, it's no longer hifi anyway).

So there are two situations when one might want regulation.

a) the input voltage is not constant or has a highish impedance

2) you want an extra stiff output voltage

How the supply reacts to a transient load will depend on how the regulation is controlled. You can easily end up with a worse response compared to unregulated.

The hypex SMPS1200 is unregulated while SMPS400 and 600 appear to be regulated for all I can tell. I wonder why.

Back to efficiency...

When the FET switches, there is little current because of the primary's inductance, except for the current needed to charge capacitances. Changing the duty cycle does nothing to the number of switching operations per second, so no change in dissipation.

The current increases after the transistor is switched on. Shortening the duty cycle will cut off this build-up of current earlier, so this means less dissipation in the transistor, doesn't it.

Looking at efficiency over load for mains powered SMPS, efficiency is poor for low loads because of the switching losses (and one way to lower these losses is to lower switching frequency), increases with load and decreases near maximum load due to the transformer and transistors not being oversized for commercial reasons. Again, this would imply lowering the output power to about 80% of max by duty cycle reduction will increase efficiency in a car SMPS.
 
When the FET switches, there is little current because of the primary's inductance, except for the current needed to charge capacitances. Changing the duty cycle does nothing to the number of switching operations per second, so no change in dissipation.

This is not correct. The dominant load the MOSFETs see is the load on the secondary. The transformer directly transfers energy; it doesn't store it (except for undesirable leakage inductance, which causes the ringing). So, as soon as the MOSFET switches on, it starts conducting considerable current. See the waveform. The top waveforms are the MOSFET currents, the bottom waveforms are the secondary currents.

The current increases after the transistor is switched on. Shortening the duty cycle will cut off this build-up of current earlier, so this means less dissipation in the transistor, doesn't it.

If there is an inductor between the rectifiers and supply caps, then it will ramp slightly higher, but it does not start at zero. With no inductor, it will spike high initially to correct capacitor sag and then level off. The attached waveform is with an inductor. Imagine how a non-switching supply would work if the AC main was a square wave.

Looking at efficiency over load for mains powered SMPS, efficiency is poor for low loads because of the switching losses (and one way to lower these losses is to lower switching frequency), increases with load and decreases near maximum load due to the transformer and transistors not being oversized for commercial reasons. Again, this would imply lowering the output power to about 80% of max by duty cycle reduction will increase efficiency in a car SMPS.

We have different assumptions going here. You assume duty cycle reduction reduces power, but I am assuming a fixed power output. Harder regulation means lower efficiency for a given output power.

When I was a car amplifier designer at Kicker, I regulated so that line regulation was maintained to a battery voltage of 11.5V. I wanted regulation for a more solid bass feel. Yes, attention must be paid to supply stability at all audio frequencies, but it's not that difficult. Regulation keeps supplies solid up to a certain frequency; capacitor stiffness does the job above that. Make sure ripple spec is not exceeded and there's no resonance or instability of the rails at any frequency. It was not that difficult.
 

Attachments

  • Waveforms-of-transformer-winding-currents-in-the-push-pull-configuration-a-Primary-and.png
    Waveforms-of-transformer-winding-currents-in-the-push-pull-configuration-a-Primary-and.png
    55.2 KB · Views: 217
Last edited:
Russell, thanks for bearing with me. I didn't realize things are so different when you drive with a square wave and have a pretty small core.

I suppose we are on the same page when we assume same output power, transistors, transformer. Then the same power has to be transferred in a shorter time, requiring higher current and hence higher resistive losse.

As you are a former Kicker designer, may I pick your brains on a few other questions that have cropped up?
- why does pretty much every manufacturer but Kenwood use toroidals when ETDs are easier to wind and probably as efficient?
- what is the philosophy on completing turns?

This here is a rare example of turns all ending cleanly in the PCB
Raveland XAB 1200 - AMP-Performance

This may mean that there are half turns but I think they took care to take the wires to the same distance from the core where they enter the PCB so the loop is slightly enlarged but closed nontheless.

Much more often you will see a turn wound to the upper end of the core (maybe because a half turn was needed to achieve the intended voltage) and then folded back onto the PCB. Since the wires carry curent and are a quarter inch at most from each other, I think their fields will partially cancel, rendering the half turn mostly ineffective and adding to resisitive loss.

This is one of countless examples of this practice:
Blaupunkt GTA 4100 - AMP-Performance

- and in keeping with the question of ubiquitous toroidals and folded back wiring, why is there one single amp topology (PNP/NPN doubled long tailed pairs with no degeneration and no emitter current sources, dual VAS, double emitter follower output stage) that is immensely popular while you find much more variation in commercial Hifi amps and AVRs? One would think that a single LTP with emitter current source would give better supply rejection, and with the addition of a current mirror better performance for essentially the same parts count. Similarly, a triple EF would give better performance, and a CFP better efficiency. Was there one milestone amplifier that defined the go-to topology for car amps?
 
Last edited:
Why is there one single amp topology (PNP/NPN doubled long tailed pairs with no degeneration and no emitter current sources, dual VAS, double emitter follower output stage) that is immensely popular while you find much more variation in commercial Hifi amps and AVRs?

Probably because it's cheap, easy to build, and don't take up much space on the PCB. Other market segments can experiment to their hearts content, there's more latitude to give people something different to buy there. Not that car amps weren't experimented on circuit wise either, just that those types of amplifiers don't make up the majority of a given company's revenue. They generally get relegated to the 'niche' category, because of their low production numbers and associated high cost. I don't think there's much demand for different car amp circuits, but there is demand for how an amplifier looks in your car. That's why you see multitudes of different amp cases with the same exact board inside it.
 
As you are a former Kicker designer, may I pick your brains on a few other questions that have cropped up?
- why does pretty much every manufacturer but Kenwood use toroidals when ETDs are easier to wind and probably as efficient?
- what is the philosophy on completing turns?

That's a very good question. Back in the mid-90s, I would have told you it was because toroids have less loss and EMI. Then I had some discussions with Robert Zeff of Zapco (if you remember his excellent amplifiers), and he held the view that E-I cores were less lossy, and had lower leakage inductance, and less capacitance between windings. He demonstrated that for me on his lab bench, so I had to concede that argument to him.

Ultimately, I kept using toroids for three reasons:
1) They had lower profiles, so they fit better in shorter cases.
2) Market perception was the toroids were better.
3) I already had a lot of experience with toroids.

So, you see, I don't really have an argument against E-I cores, except I had inertia on toroids at the time. #1 and #2 are why toroids still dominate, I suspect.

Badly designed transformers destroy switching FETs, and/or are grossly inefficient at high power. I no longer had to iterate my toroid core choice and windings, because of experience and refined design equations. I start with the acceptable power loss, and I use black-body radiation equations to choose core size. Then the B-I curves of the ferrite material to choose number of primary turns, and the desired power output to choose number of secondary turns. Choose the largest wire gauge that wil fit the window. Don't use that gauge, but use the equivalent cross-section in 24 - 30 AWG, to avoid skin effect.

Much more often you will see a turn wound to the upper end of the core (maybe because a half turn was needed to achieve the intended voltage) and then folded back onto the PCB. Since the wires carry curent and are a quarter inch at most from each other, I think their fields will partially cancel, rendering the half turn mostly ineffective and adding to resisitive loss.

I think I disagree with you about the fields canceling. Toroids were very demonstrably an integer number of turns. You either went through the hole in the middle, or you didn't. E-I cores will allow you half-turns, as you say, but if you don't go through the window, you don't couple with the other windings. Fields won't significantly cancel in the air around the transformer.


- and in keeping with the question of ubiquitous toroidals and folded back wiring, why is there one single amp topology (PNP/NPN doubled long tailed pairs with no degeneration and no emitter current sources, dual VAS, double emitter follower output stage) that is immensely popular while you find much more variation in commercial Hifi amps and AVRs? One would think that a single LTP with emitter current source would give better supply rejection, and with the addition of a current mirror better performance for essentially the same parts count. Similarly, a triple EF would give better performance, and a CFP better efficiency. Was there one milestone amplifier that defined the go-to topology for car amps?

I have not really followed what current car audio designers are doing. That is a valid design, I suppose, but it's not how I did it.

I used a single degenerated input pair, tail current source, EF buffer before the VAS, and current source load on the VAS. Lots of emitter degeneration on all of that. I also used triple EF outputs. Those are harder than they look. Cascaded EF stages have an inherent high frequency instability at high current. Curing that is well-documented, now, but I had trouble with it at the time. I got 0.01% THD, but if Doug Self's book had been available then, I could have got it much lower. I had learned much of what he says in his book, but not all of it.
 
Rusell, thanks for your comprehensive reply and sorry for not getting back sooner.

On the half turns, maybe I wasn't clear enough. I finally found a picture here Phase Linear PB 4 - AMP-Performance that contains the good and the maybe not so bad (cutout of the transformer attached below).

On the left hand side, you can see what is likely the secondary that is done right. The wires come out of the PCB, go over the top of the core and go through the hole. The last winding appears underneath the core and goes to the PCB with a minimal bow. It does complete the turn, though.

On the right hand side, near the input buffer electrolytics, you can see that the last turns of each secondary appear underneath the cure, go all the way to the top and are then folded back into the PCB.

Based on some other pics I no longer have handy, I though these were half turns, but it is really a full turn and then some extra (probably because the remove the insulation on the strands before they start winding the core). If the field is really pretty much contained by the core, this extra air turn will do little harm. The fields of the wires carrying antiparallel current will pretty much cancel so all that is less a little extra resistance.

I have seen other examples where the first turn is already too long and folded back and so is the last, and these foldbacks are not always as narrow as they are here.

I was intrigued about the triple EF and don't remember Bob Cordell or D. Self saying they are particularly critical. You need to include a base stopper or a ferrite bead but that should be standard practice also with fast double EF.
 

Attachments

  • PhaseLinear_PB4.jpg
    PhaseLinear_PB4.jpg
    144.2 KB · Views: 136
I found that triple EF loaded the VAS stage a lot less and resulted in much lower THD. That was with TIP35 and TIP36 transistors in the 1990s. With today’s extended beta devices, it’s probably not necessary.

I think they still help in my topologies. I use folded cascode now, and it’s important to keep the impedance before the output stage very high or OL gain is killed.
 
So we agree that this extra fold if narrow does does not degrade efficiency except for the unnecessary but small resistive loss?

Extended betas were around in the 80s and 90s but maybe too expensive for car audio. In fact, some of the fastest audio power transistors were made by Toshiba and Sanyo and discontinued really long ago (they were unavailable already in the early 00s). The way I read Bob and Self (and maybe I need to reread them), a triple will be beneficial even with extended beta. My question was more about the HF instability at high current. My understanding is that this is not limited to triples and the fix was well known before both of these books came out.
 
There is no loss to the extra fold except a little extra resistive loss. There would also be a bit of extra leakage inductance, but probably it is negligible.

You're correct that those devices existed in the 90s, but I was designing car stereo amplifiers then, and they were very cost sensitive. I was doing the best I could with a stringent budget. That problem kept getting worse until I could no longer design products I found acceptable with the budget I was given. The market became commoditized, as they say.

So I got out and started a career in IC design--mostly class D chips for Texas Instruments, but also a couple of AB headphone amps in all CMOS processes. On ICs, you can design circuits as complicated as you want. You're still paying almost entirely for the area of the power output transistors. I'm still doing IC research for a Taiwanese manufacturer, but no audio at work, these days.
 
Ah, I see you have a comment about triple EF stability. I didn't realize you were asking for comment about stabilizing triple EFs.

Yes, those are not always easy to stabilize, especially as the signal approaches the supply rails. Cob increases as Vce drops. Also, quasi-saturation creeps in with some devices. In simulation, I do stability analysis at many points across the signal swing, and also with complex loads, so I can jack with the voltage/current phase in the output devices. I stabilized in those days by limiting the current gain of the middle EF stage, because base stoppers reduced the BW of devices that were already low BW.

EF stages don't like to be driven by an impedance that increases with frequency. EF stages have an output impedance that increases with frequency, so EF stages don't like to be driven by other EF stages. But as you said, techniques for stabilizing cascaded EF stages are well known.
 
Last edited:
It's a simple local feedback. Here's a picture of it.

The 49.9 ohm resistor sets the bias current of Q24 to about 15 mA. The ratio of R21/R26 sets the maximum current gain of the stage to less than 50. The emitter resistor also provide some damping. I'm not actually sure whether it's the damping or the gain limiting or some combination that is giving the stability. I should investigate that.
 

Attachments

  • driver_current_gain_limit.png
    driver_current_gain_limit.png
    29.4 KB · Views: 75
Interesting. Thanks!

This is the first two EFs of a triple, right? Is there a larger value shared resistor between R26 and its counter part on the PNP side?

I suppose Self had us all convinced that shared emitter resistors on the drivers in double EFs are such a good idea that triples automatically use shared emitter resistors on both the first and second stages.

Well, R26 is sort of a base stopper for the third stage, isn't it?

I suppose if you kept R26 but connected R21 directly with its counterpart, you could investigate how much is down to feedback and how much to the damping / base stopper action of R26.

Good discussion, learning new tricks, thinking new thoughts!
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.