High end Switch Mode Power supply for Audio

I am in the process of doing a high-end SMPS design for Audio power amps. Specifications should typically be 800W capable at around +-60V. I have gotten some great help from a member on this forum already but thought a collaborative effort might work better still. One of the main goals is to have a high switching frequency (say 200KHz or better). Being technology oriented, it should ideally use state of the art components as well :)

To the non-believers, I have heard a power amp with a low-tech SMPS (operated as low as 50KHz SIC!) and non-impressive amplifier stage. You should consider trying before dismissing :)

Configurations considered to date:

1. polyphase controller designed for PC motherboards -- ideally full 12 phase setup :). This would probably yield the "strongest" supply -- as these in 4 phase mode yield 90 Amps for CPU power. The apparrent problem is that the output voltage sensor is optimized for low to very low voltage and this troubles me. I am seriously considering building this (transformer based) with magnetics (slightly modified), semiconductors and glue logic as per the Infineon reference design below (as distinct 2 transistor forward converters using high side signal to switch). To date it appears that www.linear.com have the "best" controllers. Would require separate PFC controller. With this approach, just about any frequency can be programmed up to 1MHz or so. Still 200KHz per phase at 12 phases would be very very cool indeed.

2. ZVT design. This is more complicated (theory of it), particularly as switching losses can go high when power drawn is low. Again, I have found an interesting chip from Linear Technologies http://www.linear.com/prod/datasheet.html?datasheet=601 which does simplify design, particularly balancing of current. Main problem seems that this type of design is sensitive to output load. If we could base this on an actual Linear design that would be good even though I am not all that chuffed about the gate drive circuitry that Linear uses (transformers everywhere). Still this is the most modern topology and yields less EMI.

3. Infineon's 200W reference design http://www.infineon.com/cgi/ecrm.dll/ecrm/scripts/public_download.jsp?oid=13467&parent_oid=-8681. This is very very cool. Way to go German dudes! Simple 2 transistor forward converter. Problem: It is simply too small. I have solved scaling the PFC section, but am struggling with the power transformer, reason being that larger pot cores do not seem to be available in low loss materials required for high operating frequency. Still, I am considering taking a multiple of the output transormers (say 4) and using 4 paralell sections of wire with a quarter of the primary turns per unit -- all wired in series. Secondaries with foil wiring to suit. Now, all I have to do is reduce the current sense by the same factor and I verify that switches are capable of driving the thing.
Based on the quality of the reference design document -- and the usage of state of the art components just about everywhere this should be the simplest way to get there -- if I am doing the magnetics work right. Still, larger transformers would look a little neater ... One other problem ... Epcos have REALLY LONG delivery times and minimum orders of 60 cores which is a little painful. Cores are inexpensive, though.

Learnings to date: My understanding is that Infineon are basically the switching FET and diode masters of this universe (pricing of FET's are most reasonable (Euro 1.5 each), diodes are highly priced (Euro 8 each) but they are single sourced and truly radical). Linear Technologies seem to have just about the best regulator chipsets. www.mag-inc.com have some great cores, particularly well suited to PFC section. Pot cores seem to have maximum shielding but are problematic to get in large form factors. Toroids are probably not optimal for power transformer. Winding the transformer, secondary might well be best foil wound. Planar magnetics (if available) might also be very useful, particularly if we get into a ZVS design. Synchronous rectification considered but believed to be primarily for very low voltages. Current doubling outputs still considered.

How about it guys, anybody have deep knowledge in magnetics design for high frequency SMPS???


Try: http://www.keisoku.co.jp/pw/scatfaq/index.html

At the bottom of the page you can find a "Tesla converter" by Dr. Cuk (modeling by Shin Nakagawa). The conversion efficiency is about 95%, which should be close to state-of-the-art. More circuit data is here: http://www.keisoku.co.jp/pw/scatfaq/tesla.exe

The circuit is quite new, and will be the topic of an upcoming feature article in "Transistor Technology", one of Japan's better-known EE magazines.

If you find it interesting, I may have another article or two on this Tesla converter that I could upload.

hth, jonathan carr
I have checked the link out, and it was very very interesting. Thanks! I am kind of leaning towards doing either:

a) Polyphase with Infineon output stage and magnetics
b) Multiple transformers on the Infieon reference design for a forward converter.

That is of course unless the Japanese come up with something totally radical which is quite possible! If they do, I hope it is of a type which is not sensitive to output load to function.

Please keep be informed of developments, Jcarr

Slew rate

search on Linear (www.linear.com)'s site for "low noise" switchers and drivers. I will caveat any comments here by saying that I don't design SMPS power supplies for a living, but I have built many and consider it the most intriguing and potentially elegant application of the DIY craft. Perhaps it's just my foibles, but I have found Linear and Nat Semi to be most supportive. Read the ap-notes at Linear and you'll get the message that it's all about a little calculus with the slew rate.
I am curious as to what your design goals are, low ripple, high efficiency? I have nothing against switch mode for power supplies. My concern would not be the switching noise, so much as the loop reponse having a response well into the audio band. Increasing the switching frequency will certainly help. Multi-phase controllers can certainly help if you have a good controller.

My main concern with switch mode supplies is the jitter they can create in digital equipment. To that end, the Multi-phase structure will significantly cut down on your input ripple and hence effect on other audio pieces in your system.

For AC\DC supplies, you should probably look at ST also for FETs, and Controllers. ST has some nice PFC correction chips. It's FETS are top notch as well. Fairchild also has very good FETS especially with all the acquisitions they have made.

Very interested in how things go!

I am basically considering anything under 200KHz unsuitable by default (unless it is multiphase).

Check out the app note from Infineon (200W PSU) and see for yourself how cool it can be. Now imagine running a 12 phase unit in that mode ...

I have found large cores from Epcos now so I am kind of thinking that using the learnings from that application note and possibly considering another controller (for ZVS say) with the larger core HF transformer would be suitable.

My design goals are to create a low impedance power supply that interfers as little as possible with the sound. That means putting the feedback below the audio band ("strong psu" that inherently does not drop + large holdup capacitance), or above the audio band (fast feedback, also with large holdup capacitanc ...).

I will be looking more into this soon. Thanks for all the great responses so far!

I am currently working on a switching power supply for an audio amp of the same power range. It consists of an off-line ZVS active PFC front end with a ZVS converter stage. The secondaries include current doubler topology with synchronous rectification. Pot cores are the wrong direction. Pot cores are not a good choice when using high currents. In this case it could be about 10-20A secondaries (max). Pot cores are not good for this application because of the increased copper cross section needed to carry this current and the limited amount of "opening" to allow wires to escape from the core. If the primary is around 400V from a PFC then arcing to the core is most likely and undesirable. The use of current doubler and synchronous rectification extends the ZVS load range into lower current loads and thus increasing efficiency and performance over the entire range. For those less knowledgable about Zero-Voltage Switching, the transistors in a full bridge arangement are controlled via phase-shifted gate drives and allow the transistors to "turn-on" when zero voltage is across them (due to parasitic components utilized within the MOSFETS). Because there are no volts across the transistor when it is turned "on", there will be nearly zero noise in switching. However, this effect is dependant on the amount of current which flows in the primary and under light load conditions there is not enough energy to place zero volts across the transistor. Thus a leakage spike will occur. But it will be small because there are not many volts across the transistor and there is not much energy in the cycle. So the trade off at light load is easily made for the benefits at full load current. I highly recommend the Unitrode series of controllers from TI. They are outstanding performers. I also recommend MOSFETs from International Rectifier and cores from Magnetics Inc. available from Allstar Magnetics on the web. The frequency of such a ZVS supply will be around 150kHZ to 200kHz because it is a quasi-resonant topology. Also it is recommended to use magnetics amplifier post-regulators for the auxiliary outputs which are not directly in the feedback loop. Current mode control is stongly advised!!!! Would you like me to go on? This message is rather lengthy......

Please do go on.

However regarding current I am not sure that I agree with you. The Infineon application note has as I recall 20A at 5V and 8A or so at 12V from one small pot core using tape secondaries -- all from a 360V primary. The pot cores used have pretty good opening slots.

The pot core I am considering using has 10 times the mass and is made of the same material.

I have looked at Unitrode controllers but find the application notes lacking compared with what to me is the reference to date (Infineon). Regarding transistors, I am sure IRF make great devices and have standardized on diodes and transistors from them. However, it was Infineon who changed the HV on resistance from quadratic to linear function of voltage and they still appear to be way behind Infineon in this particular area. They are also driving the field with diodes. Infineon uses a PFC core from Magnetics made out of KoolMu and i have been using magnetics from them for years. Very good company, very helpful as well.

As I said, please do go on. If you can provide us with information, app notes, ideas, guidelines, device numbers etc. I would be particularly chuffed.

Thanks so much. I will re-read your posting tomorrow to attempt to get the full benefit of your contribution.

If you are going to try a zero voltage switching supply it is beneficial to use a MOSFET which is Coss characterized. Since this parasitic output capacitace is such a major player in the performance and operation of a ZVS converter, it must be well known. This parameter changes drastically with a changing Vds. And because of this it is often estimated by an 8/3 factor for a full bridge topology. International Rectifier has specially designed devices which have an "effective" Coss value which can be used as a constant while rising from 0V to 400V Vds. Also for a ZVS you want a low gate capacitance and a robust VDS rating with high rise times and a low Rds(on) value to minimize heat dissipation. I am only semi-familiar with the Infineon devices but as an engineer I tend to stick with the IRs because they are very robust, low cost, and readily available. I have had less desirable results from other brand products...(never used Infineon). For an active PFC the typical core type used is a high flux Molypermalloy Powder core (High Flux MPP). This is because you want a higher permeability leading to higher energy storage at high frequency. KoolMu cores are less likely to be used for the application you originally described. I am skeptical of a data sheet which lists this type of material for this frequency and power range. As for the pot cores, I was just mentioning the ruled of thumb when considering a core geometry. You must also keep in mind that: increasing frequency, increasing core mass, high currents, and increasing flux density contribute to increasing transformer losses and must be considered. The RM and PQ geometry cores are designed to optimize core volume and surface area to minimize temperature rise of the transformer. I can also recommend toroids for their low cost and low EMI due to a completely closed magnetic loop path. If high currents are to be used in the primary it may be necessary to "gap" a particular core in order to "stretch" the B-H curve and allow it to operate at higher currents and low flux densities. For the output filter inductors, if you choose to use them (current fed topologies eliminate output inductors for a single inductor before the full bridge) then you can use MPP or KoolMu for them. EE and EI cores will most likely require gapping. Keep in mind that you can accurately gap the two outside posts half each or gap the center post twice the amount. But, by gapping this type of core or any core, you are allowing the EMI to radiate through the air medium. Therefore, gapping the center is better than gapping the two outside legs from an EMI standpoint. Cores can be ordered with a particular length of gap for you application. Also, make sure that your full bridge transistor gates are transformer isolated because the source terminals will be at the same potential of the DC input, and thus if a PFC is used could be 400V!!!. Your secondary feedback must also be feed back in an isolated form, either through an optoisolator (most common) or through a transformer. As I stated previously, current mode is strongly recommended and a low loss, high frequency current sense transformer is the most likely method to be employed for this method. If you are not familiar with current-mode control, it is a method to monitor the current for each half leg of a full bridge so that the transformer is subjected to equal and opposite volt-second products. If this is not done, then the operating point of the transformer will "walk" up the hysteresis loop and eventually hit the saturation region where you transistors will drive a seemingly short circuit and explode!. This flux imbalance can also occur by unequal drops across series transistors in the bridge. Example would be if an upper MOSFET had an on resistance of say .1 ohm and the corresponding lower MOSFET was .102 ohms there would be a DC bias and each cycle would slowly start walking up the hysteresis loop because of unequal IR drops. The on resistance is highly dependant on temperature so it is difficult to know and control. Current-mode control corrects for this on a pulse by pulse basis ensuring operating around the origin. I can answer any questions you have on the subject. A switching power supply is not an easy design at all. And the hard pard doesn't come until you have the thing built. The debugging and loop stabilization is the trickiest.....


I am investigating a SMPSU too for powering a two channel 50W class-A amp. I need app. 250W of steady power. I am considering using a PFC first to make 400V DC (not that difficult) and then down-convert it separately for each channel to + and – 35V with a straightforward forward bridge converter in open loop (at near 100% duty-cycle). Or maybe a half-bridge forward converter. I have to dig this out further. This can be build small and efficient.

I am thinking about giving the PSU its own housing, so filtering can be effective.

Any thoughts about it?

Just opened an account on this website. I was just reading the traffic on the site. I stubled into this site today while searching for some datasheets for a Toshiba part number.

I think I'm going to like it here! I am also a fellow high power audio fanatic, and a high power SMPS designer. It's interesting listening to the traffic on off line switchers.

I don't have any high power high input voltage experience, but if anyone is intersted in playing with low input voltage (12V, ie Automotive voltage) high power switchers, let me know. I currently have a mostly finished 2000W 12VDC to +-65 VDC (from scratch design) switcher staring me in the face. Next to it is a 450W AVG/Channel (4Ohms) 4 Ch amplifier (also completely from scratch design), and also mostly finished. Yeah, I like car audio!

As for the traffic so far, My experience is that high frequency is not really needed, and can play havoc with ShortWave Radio Receivers in the area. 40 to 50Khz switching should work fine. Any higher frequency, and you start having problems with the switching transients on the Power switches (At least, at low voltage input). As stated in an earlier post, High frequency usually means high losses in the switching device. To combat ripple, add a lot of output capacitance. With gobs of output filtering closed loop response can be a little sluggish with no adverse affects on the sound output. Even if you get some low amplituce ripple on the rails, a good audio amp should be immune to it. (Think of what a couple of hundred watts does to a 50/60 Hz linear unregulated supply)

Anyway, looking foreword to reading any future posts.

I disagree with the previous statement on no requirement for high frequency operation.

In my opinion, the feedback loop should be well below audio frequencies or well above audio frequencies. That means that you either need to draw constant current in the output stage for the first scenario, or have 20KHz*6-->10 switching speed in order to be able to stay above the audio band.

On the other hand, I have not build anything like this yet so take all this with a pinch of salt.

Some comments on switching frequency.

A higher switching frequency allows a higher looop bandwidth for the regulation of the converter output. A common rule of thumb is to make the loop bandwidth at least a factor of 5-10 less than the switching frequency. Not only does this make the loop easier to design and analyze, but it also means that the loop is not significantly affected by the ripple on the output.

In theory, the key parameter of interest regarding the amount of output capacitance is output impedance, which should be very low for good load regulation. This can be achieved by lots of capacitance (same as with a conventional power supply), or a good control loop, or both. If you use a lot of capacitance, you can use a lower loop bandwidth for a given output impedance. If you have a higher loop bandwidth, you can get away with less capacitance. If you do the latter, don't forget to consider capacitor ESR. All kinds of tradeoffs to be made here!

If you don't use ZVS or some other type of soft switching, keep the switching frequency low.

I don't know what effect any of this has on the sound. I haven't ever designed a commercial supply, only research supplies, so you can take what I say with a grain of salt.

About PFC:

Is there a reason that you want PFC? Do you need it? I realize that in theory it is a good idea, fewer line frequency harmonics, etc., but is the added cost and complexity worth it? For a relatively low power supply, it may not be. It might be worthwhile to design a converter without PFC first. You can always add it later, although it will probably require some redesign.

High frequency...

In my experience, High frequency looks good at first. The problems start to arise when actually building/testing a supply.
I, like you firmly believed high frequency was the only way to go. However, after spending enough time trying to achieve high frequency, I bit the bullet and lowered the frequency. After testing the supply with the amps, I found I had been worrying about nothing. The rail never moved when running actual audio through the amp. Even at high power levels into a dummy load.
(This was probably due to the fact that a lot of the power was consumed bass frequencies)

Keep in mind, I play with with 150+ Amps or of input current, so you may now run into the same difficulties as I have. (We'll be trading grains of salt here ;) High di/dt waveforms and the tranformer leakage inductance can generate enough voltage to take out a mosfet fast. I suspect depending on your topology you may run into smilar problems.

SMPS Soft switching and slew rate

So I will paraphrase the apnote from Nat Semi http://www.national.com/an/AN/AN-1229.pdf since it doesn't seem to be getting any traction:<p>One of the reasons for which DIY'rs have difficulty with switch mode power supplies, particularly when they go over 50kHz -- is that they fail to take into account the "criticality" of lead length. <p> As I said earlier, you have to go back to your first E&M class , V = L di/dt -- each tiny little trace has an inductance of about 20nH per inch. di/dt isn't the switching speed, it's the transition speed (and all of those lovely rich harmonics!) which is much faster. If the inductor , in this case "trace", is storing energy it takes a finite period of time to release the energy leading to lowered efficiency. The folks at NatSemi are pretty emphatic that via's (for connectivity purposes) should be avoided for this reason <p> With respect to ground planes -- double sided is best as a ground plane can soften transients and reduce EMI. NatSemi even gives you "X Marks the Spot" diagrams for where to attach the ground plane the circuit on the reverse side. <p> It's a good ap-note, certainly not written by a liberal arts major, (as a former editor I cringed at some of the sentence construction) but one I recommend.
Lighter And Cheaper

Take a good look at modern television and monitor mass produced smps stages.
TDA and STR prefixes are chips that are cheap and easily available - go google.
The transient variation in load for audio is different to tv applications, but modern mass produced smps techniques seem pretty good.
High power car audio relies on smps operation and is another area worth investigating, and is done reliably and cheaply on a very large scale, especially in modern cheap asian car audio amplifiers.

With apropriate filtering, shielding and grounding techniques, I think smps are valid and acceptable for high quality, high efficiency audio applications.