SMPS for motorcycle intercom: some questions

[DaBit]

Hi guys, I am new here. Let's introduce myself a little. I am an electronics engineer living in The Netherlands who fills his days doing mostly digital board-level designs, programming microcontrollers, DSP's and FPGA's en writing some GUI test-software. Spare-time projects vary a lot; building extreme cooling devices for PC's, some electronics to solve 'things' and software.

Now, one of the electronic devices I have built is an intercom for motorcycling. The design is completely analog. Two headsets (+mic) can be connected to the device. Microphone signals are high (300Hz)/lowpass(4kHz) filtered and, due to the enormous dynamic range of the signal coming in, fed to a variable gain amplifier/compressor. Following the microphone stage is a mixer stage which mixes in other audio sources such as the GPS voice and/or music. Then, the mixed signal is again fed to a variable gain amplifier which is controlled by the incoming wind noise signal. This raises volume when background/wind noise rises. As a last step, the audio is amplified with a couple of LM386's.

Now, this works, sort of. There are a few problems such as ground loops, the very unclean power supply of the bike, and amplifier clipping when speed goes above 75mph/120kmh. I have to use isolation transformes in almost all incoming audio lines.

Therefore, I decided to go back to the drawing board. I will design/build a new one with upgraded specs. But all this starts with the power supply. I will do the power supply and headset amplifier first, and later on move to a DSP for microphone signal processing.

For the new design I would need the following voltages:
- Isolated approx. 30V or +/-15V for the headset amplifier and analog circuits.
- 3V3, 2V5 and 1V8 for the (later) digital stuff.
- Two isolated voltages of 5-15V which will be used in the input stages to provide isolated audio inputs. This to avoid ground loops.

I haven't made a power estimation yet, but we are talking about 15-20 Watts maximum. Now, the easiest way to accomplish this seems to be a SMPS using a flyback transformer. I have done a few more flyback SMPS-es using the UCC2800, but for those projects I had the flyback inductor made for me according to spec. For this personal project that solution is too expensive.

Thus, I started seaching for obtainable gapped cores for a flyback transformer, but I cannot find one which I can buy. Gapping the cores myself with a file or Dremel seems a bad idea.

Now, of course I can choose another topology instead of flyback so I can use an ungapped core. Forward converter and push-pull topologies can be implemented with few extra components, and the push-pull converter can possibly use smaller magnetics and capacitors reducing the board size.

But I read everywhere that these topologies are single-output only. Why is that? Why can't I wind a transformer with center-tapped primary in push-pull and have multiple secondaries of which I choose one for feedback (3V3 digital supply would be the prime candidate since the rest of the supplies are not so voltage-critical.

What would you guys do?

[wes-ninja250]

Have you considered using DC-DC converters (like those on ethernet cards) instead? It's not like you're needing a whole load of power.

Also, when you mention the unclean supply on the motorcycle -- have you verified this with a 'scope? Have you checked the cleanliness of the connections from your battery to the regulator/rectifier? How about after adding filtering caps nearby?

I'm wondering because normally your helmet is normally only a few inches above a powerful radio transmitter -- the CDI (or points) and plugs!

Have you changed to resistive-type plugs? E.g. if your bike takes C8HSA, you can often substitute in CR8HSA which will somewhat reduce EMI.

If you're a super-duper DSP guy, maybe you could design a digital transmission system, maybe based on GSM, which would eliminate radio noise with ECC.

Also, have you considered that the clipping at high speeds is perhaps caused by the microphone overloading? Wind noise SPL inside a full faced helmet at 75mph is often in excess of 100 dB. Which is why I wear earplugs when I ride outside of the city.

Oh -- another suggestion -- instead of going to all that work of designing a SMPS, you could possible develop a more traditional power supply, by directly tapping the output of your stator (usually three wires around 16AWG). This will yield three-phase AC, which you can turn into whatever voltage you like with a transformer. Bridging and regulating that yourself might very well yield the clean 15V rails you are looking for. If you do this, check the schematic of your motorcycle, some bikes use one of the stator phases for the headlight and the other two for charging. Most use all three for everything.

Wes

[DaBit]

Originally posted by wes-ninja250
Have you considered using DC-DC converters (like those on ethernet cards) instead? It's not like you're needing a whole load of power.

I have, but they are quite expensive. And I would need at least 4 of them. Besides being expensive, it would be quite bulky also. Much more bulky than an SO-8 switching IC, small 3C90 toroid, one or two DPAK MOSFET's and a few diodes, capacitors and resistors.

Also, when you mention the unclean supply on the motorcycle -- have you verified this with a 'scope? Have you checked the cleanliness of the connections from your battery to the regulator/rectifier? How about after adding filtering caps nearby?

See 'unclean' as 'normal for cars and bikes'. There is a 50Amp generator charging the battery, ignition, fuel injection, etc. And the only device capable of creating a DC-like voltage out of that is a small, years-old battery. That is not a good recipe for clean power.

I did not measure it with a scope, simply because I don't have electricity near the bike.

Have you changed to resistive-type plugs? E.g. if your bike takes C8HSA, you can often substitute in CR8HSA which will somewhat reduce EMI.

Of course.

If you're a super-duper DSP guy, maybe you could design a digital transmission system, maybe based on GSM, which would eliminate radio noise with ECC.

Electrical noise can be eliminated by using proper filtering. My currrent intercom uses a 680uH coil and a bunch of capacitors to filter the incoming +12VDC, then it is stabilized to 10V using an 7810. Negative opamp power supply is done with a charge pump and 79L10. This works good enough; I don't suffer from ignition/generator sounding through. Unless I create a ground loop.

I have an algorithm worked out partially which uses spectral subtraction to remove wind noise. This is an easy algorithm; incoming signal is Speech + Noise. If we subtract noise, the remaining component is speech. In the time domain this won't work due to phase shifts, but subtracting magnitudes in the frequency domain works fairly well. Wind noise is reduced with some 30dB. The only problem with the algorithm as it is now, is that it adds some kind of melodical noise. Sounds a bit like someone blowing over a couple of half-empty beer bottles.

Also, have you considered that the clipping at high speeds is perhaps caused by the microphone overloading? Wind noise SPL inside a full faced helmet at 75mph is often in excess of 100 dB. Which is why I wear earplugs when I ride outside of the city.

The microphone _is_ clipping. That electret thingie produces more than 3Vpp at it's output . I will replace the electret with a dynamic 'microphone'; the speaker out of a cheap walkman earpiece. Wind noise also works on both sides of the membrane, which already provides a bit of wind noise removal.

But music is also clipping, which is end amplifier related.

Of course I wear ear-plugs too. I drive about 35000km / 22000mi a year. Without ear plugs I would have suffered serious hearing damage.

My ear plugs dampen incoming sound with 34dB (low frequencies a bit less)

Oh -- another suggestion -- instead of going to all that work of designing a SMPS

It's not my first, although I do not have THAT much of experience, I only used the flyback topology, and I had the magnetics wound for me.

If you do this, check the schematic of your motorcycle, some bikes use one of the stator phases for the headlight and the other two for charging. Most use all three for everything.

Mine rectifies all three. But this solution is not acceptable; I want to be able to just plug it in the socket, and use it on another bike too.

Thus, the question remains: small push-pull or forward converter? I am leaning towards the push-pull. And why can't there be multiple output windings on a push-pull transformer? Seems a bit unlogical to me.

[Eva]

Multiple outputs are also achievable with push-pull transformer-coupled buck converters. There are several approaches:

- The simplest way is to use one secondary winding with its diodes and one buck inductor for each output voltage. Output voltages will tend to follow turn ratios, but cross regulation will be poor, though.

- The next approach is to use also several transformer windings and diodes but with a single coupled inductor whose turn ratios shoud be similar to the ones of the transformer. Output voltages will also follow turn ratios but cross coupling will be dramatically improved.

- The next approach is to use a single secondary winding, with diodes and an inductor for each output. A saturable inductor is inserted between the diodes and the buck inductor for all but one of the outputs, and each one is biased with a variable current whose value depends on voltage error for that output. This allows to delay the switching waveform, thus proviging independent duty-cycle adjustment for each output. Very good cross-regulation is achieved at the expense of complexity with this approach.

AT and ATX computer power supplies use the second method in order to get +12V and +5V, and most ATX units use the third approach to derive +3.3V from +5V secondary windings of the transformer.

Anyway, at low oputput power levels a flyback converter is simpler and more convenient.

[DaBit]

Originally posted by Eva
Multiple outputs are also achievable with push-pull transformer-coupled buck converters. There are several approaches:

- The simplest way is to use one secondary winding with its diodes and one buck inductor for each output voltage. Output voltages will tend to follow turn ratios, but cross regulation will be poor, though.

This would do for this supply. Digital supply should be tightly regulated, but amplifier supply and isolated audio input supplies are not that critical. And this would mean off-the-shelf Buck inductors

Anyway, at low oputput power levels a flyback converter is simpler and more convenient.

Very true, but I am worried about gapping the core myself. I checked a few suppliers, but they won't sell me pregapped cores in small quantities.

I haven't decided on the core yet, but an EFD20 or super-standard ETD29 core, 3C90 or N87 material, would be a good compromise between availability and size. Both are overkill for the application

[Eva]

It's better to have the gap in the center leg only because less stray magnetic field is produced, but gapping all the legs also works fine and it's done in a lot of equipment.

I use sheets of non-conductive and non-ferromagnetic stuff with known thickness, like paper, plastic, PCB fiberglass and wood, to create gaps.

[DaBit]

Originally posted by Eva
[B]It's better to have the gap in the center leg only because less stray magnetic field is produced, but gapping all the legs also works fine and it's done in a lot of equipment.

I can live with a bit more stray magnetic field. A piece of sheet metal between 'sensitive' audio stuff and the PSU does wonders.

How does a gap X in all three legs of an 'E'-core compare to a gap Y in the center leg? Thus, let's say I need a 1mm gap, but I decide to gap all three legs. How wide must I make the three gaps to equal one gap of 1mm? I'd say 0.5mm...

[Eva]

Assuming the sum of the cross-sectional area of both side legs equals the cross-sectional area of the center leg, an air-gap of X mm in the center leg only acts like an air-gap of X/2 mm in all the legs.

If you don't have enough data about the core and the material, you will have to build test inductors and increase air-gap length until the target I^2*L energy storage capability is achieved. A larger air-gap reduces permeability (and inductance given the same number of turns) but increases saturation current and I^2*L product (energy storage formula is actually E=.5*L*I^2).

[DaBit]

Originally posted by Eva
Assuming the sum of the cross-sectional area of both side legs equals the cross-sectional area of the center leg, an air-gap of X mm in the center leg only acts like an air-gap of X/2 mm in all the legs.

And what if that is not the case? Is it then just like a series/parallel circuit of capacitors? But this time with cross-sectional area?

Is the assumption you made valid for EFD and/or ETD cores?

If you don't have enough data about the core and the material, you will have to build test inductors and increase air-gap length until the target I^2*L energy storage capability is achieved.

That's always possible, but I'd rather do the mathematical homework before experimenting. Saves time in the end.

[Eva]

I don't know how to calculate it for other cases, but it may be probably a matter of keeping the air-gap volume constant. Anyway, most cores fulfill the S_sides=S_center criteria because this is the optimum shape, including ETD cores.