Class-D Output Filter Designs, SURE TDA7498 as example

[trevmar]

A year ago, when I set out to bring myself up-to-date with Class-D amplifier designs, this DIYaudio forum was a great help. It has been 35 years since I last researched and published Class D and Class E output circuit designs, and the new audio power amp chips were exactly what I needed for a new system I am building, based around a MiniDSP and my bi-amped "ground effect omnidirectional" speaker design.

But one thing which was lacking here was detailed information on available output filter technologies and components. It seems as though much of the technology produced in the '70s has disappeared into the mists of time. So I decided to document my deliberations, and make the design available for all DIYaudio members.

I couldn't buy any 'reference' TDA7498 circuit boards, so I bought a few SURE Electronics boards from Ebay, ripped out the key components, optimized them, soldered in the latest surface-mount TDA7498 chips from Mouser, and started working on today's design issues.

The previous threads I have started here have gotten quite messy in the handling of images, so I decided to put up the complete 'tutorial' on my personal website, and duplicate any key material here. I would also be happy to answer the technical questions that members may have.

I started by reducing the SURE board's distortion at high frequencies by ten times. From 3% to 0.3%. By changing just two components on the board. But you can see on the photo below the full set of changes I made to the SURE assemblies before I was happy with them

The result has been to make my SURE TDA7498 assemblies into top-notch (IMO), clean, powerful audio amplifiers. More than adequate for the compact HiFi speakers they are intended to match.

The leftmost attachment is the (final) modded board (without the heatsink) with the measured 8 ohm THD vs frequency curve next to it, followed by the distortion vs power-out curve. Note that I have separated the second, third, fourth and fifth harmonic distortions on the power-out curve to make it easier to see the interplay between even and odd harmonics.

The complete tutorial is on my website at Trevor Marshall - Class D Audio Amplifier Design - TDA7498 Output filters

This output filter topology will work equally well with most of the new >50W bridged amp chips now in the market.

Happy New Year to one and all....
.

[vacuphile]

Very interesting, thanks for posting this.

vac

[trevmar]

Glad you found it useful. I have now updated my webpage with the MultiSim simulation file, so you can replicate my results, and explore the design issues in more detail. The schematic is attached below.

I have placed 0.01 ohm resistors in series with the inductors to allow me to plot the inductor (and FET switch) currents. There are three Bode plotting instruments, XBP1 for the inductor current, XBP2 for the output voltage vs frequency response, and XBP3 to show the voltage across the damping resistance, and hence allow its power dissipation to be calculated. A compromise has to be struck between power dissipated in this resistor at high frequencies, and the damping applied at the filter resonance. In this case, I decided to ensure that the peak current was just within the capabilities of the FET switches in a good TDA7498 (7 amps typ.)

The direct link to the simulation file is:
http://trevormarshall.com/class-d-tu...ut_filter.ms11
.

[jyoung070848]

I find your work very interesting. I think you show that the output filter is the greatest problem with class-d amps and is the reason reviews of class-d amps vary so widely. The sound changes with different speakers more than almost any other amp due to the output filter. Your dummy load graphs show how the highs vary with load, and this is very audible. I am ignorant of how changing the capacitor type can cause the distortion to go down. Do you know what the mechanism is? I can hear the difference between capacitor types (subtle, yes), but some people see the whole thing as "snake oil".

Would you do me a favor, and honor, by looking at my recent post entitled "Class-D Output Filter - A Case Study". My aim is to show how DIYers can greatly improve their class-d amps just by designing the output filter to match their own speakers. It is fairly easy, you only need to measure the DC impedance and inductance of your high freq driver, and solve the equations.

[Osvaldo de Banfield]

Very interesting thread !!!

Thanks!!!

[jyoung070848]

You may be able to keep the class-d board inductor and let the cutoff freq change by changing the capacitor to match your speakers. If the inductor looks pretty good.
I did a rough approx in Excel and it looks like the freq won't change enough to be a problem. And I am confident that getting the Output filter to match your speakers will make a much greater difference than changing the cutoff frequency, as long as the frequency stays significantly above 20,000Hz (i.e. 35,000 Hz or so).
To get the capacitor value to match your speakers, first solve:
f=1000*(0.2251*Rs)/Lf
This gives the new cutoff frequency using the actual value of the on board inductor. Remember to change the value of the inductor from uH to mH for the equation then change it back to get the final value. If the cutoff freq is more than 35,000Hz you can keep the inductor.
Then solve for C:
Cse=1000000*(0.1125/(Rs*f))
This will give you the capacitor for the output filter (not the zobel, don't confuse the two).
You can then unsolder one lead from the on board output filter capacitor (or capacitors if the output is BTL) to disable it. And, Disable the on board zobel by unsoldering one lead on the zobel capacitor or resistor. Now you can add an outboard output capacitor(s) and zobel on the outputs with test leads and play with the values. Don't worry about the stray capacitance and inductance of the test leads. Just keep the leads fairly straight.
If your output is BTL, the capacitors you should use are:And since the T2 output is BTL:
Cbtl=Cse/2
On the zobel:
If you have calculated the appropriate zobel for your speakers using the method mentioned in the first email, you can increase the highs by increasing the zobel resistor, or decrease them by decreasing the resistor value. And tweak to your hearts consent. :-)

[trevmar]

Originally Posted by jyoung070848

I did a rough approx in Excel and it looks like the freq won't change enough to be a problem.

You do realize that I used a SPICE emulation to understand and optimize my design? Have you read what I actually did before making your suggestions about using Excel?
.

[jyoung070848]

Originally Posted by trevmar

You do realize that I used a SPICE emulation to understand and optimize my design? Have you read what I actually did before making your suggestions about using Excel?
.

I am so sorry. I accidentally posted this message on your thread before I realized where I was. The message is NOT about your work. Your work is so obviously thorough and professional I could not begin to presume.
I thank you for the tutorial and have been promoting it elsewhere. I've used some of your suggestions already and am going to compare your output filter to one I did next week. I want to see how a well designed filter that is meant to work well on a variety of speaker impedances compares to the one I taylored specifically to my speaker impedance and inductance.
James :-)

[trevmar]

James, I didn't mean to be so strong. Sorry if I offended or discouraged you. I did see your other thread about output filter design. If you have any trouble getting the emulation set up, feel free to write, and I will help you in any way I can. It really is much easier to see what each component is doing with SPICE...
.

[jyoung070848]

So far I've been too nervous about using spice since I have no experience with it. Physics is my background and designing speakers with simple programs my only experience. My approach to the same problem you looked at is to divorce the output filter from the amp and look at it as just another simple butterworth filter on a speaker with a zobel that takes the speaker inductance out of the equation. I did my first real test today and the result was nearly spot on. It actually shocked me because I expected something to go wrong. Unfortunately, today's test happened because I discovered that I made an error in how I moved the problem from btl to single ended and the results of my calculations were wrong and I had to redo them. Oh, well! I am new at this.
James