Stock SI spectrum analysis

BWRX · 2005-04-08 2:56 pm

Hello everyone. Last night I did some sonic impact testing with a dc bench supply, signal generator, oscilloscope, and dynamic signal analyzer. i used an 8ohm 20W resistor as the dummy load on the left channel, had the supply set to 12V, set my laptop to output a 1Vrms 1kHz sine wave to the SI input (verified with the oscilloscope), and hooked up the dynamic signal analyzer to the leads of the dummy load. i also set my camera up right in front of the signal analyzer and took shots of the screen for a bunch of settings. The first photo is with the sine wave off and the amp is just turned on (you know how the stock volume knob clicks when it's turned on, right after it clicks). The second photo is with the sine wave on and the amp is still only turned on. The third photo is with the output adjusted to 59mVrms. The fourth photo is with the output adjusted to 108mVrms. The photos after that are with the output adjusted in approximately 100mV intervals up to 3.793Vrms. My camera battery died or I would have kept going

http://www.personal.psu.edu/users/b/w/bww129/pics/amp/1kHzTHDtest.mov

At 1Vrms out it drew 83mA from the supply, 2Vrms - 137mA, 3Vrms - 210mA, 4Vrms - 300mA, 6Vrms - 538mA, 7Vrms - 689mA. 7Vrms was a little before the onset of clipping (volume knob a little before the 1 o'clock position). Higher output voltages resulted in a dramatic increase in current draw from the supply, and the noise spectrum from about 13kHz and up increased dramatically as well. The harmonics of the 1kHz tone were clearly defined on the screen and had a very sizable amplitude compared to the level of the fundamental.

So, from my observations so far, around 7W into 8ohms ( (Vrms*Vrms)/Load ) the distortion begins its ascent to the heavens. Anyone know how to determine THD from the series of images in the movie?

BWRX · 2005-04-08 6:15 pm

I just ran some calculations to find THD using dBV values I picked off the 2.805Vrms output photo (corresponds to approximately 1W into 8ohm load). Assuming I used the correct values (I can't be that far off), I calculated the THD to be 0.000007% at 1W into 8ohms. Assuming the noise (+N) is around the -90dBV level, the THD+N is practically the same value. Does this seem right?

I used the formula:
%THD+N=[sqrt(H2^2+...+H20^2+N^2)/sqrt(H1^2+H2^2+...+H20^2+N^2)]*100
and you just remove the N (noise power) terms for %THD.

H1=fundamental harmonic power level in Vrms
H2=second harmonic power level in Vrms
...
H20=twentieth harmonic power level in Vrms
N=noise power level in Vrms

Was I correct to assume that I can read a harmonic's dBV value from the photo, calculate its Vrms value, and use H1=(calculated Vrms value from dBV)^2/8 to get the harmonic's power level?

tiroth · 2005-04-08 6:26 pm

Something got screwed up there. I can assure you that neither your analyzer nor the SI has -143dB THD!

You could use this little javascript to figure it out easily.
http://www.anidian.com/audio/misc/calc.shtml

BWRX · 2005-04-08 6:55 pm

Thanks tiroth. I figured my calculations were incorrect because of the extremely low number :cannotbe:

h1=8.96dB (fundamental)
h2=-80
h3=-63
h4=-85
h5=-77
h6=-87
h7=-77
h8=-79
h9=-70
h10=-75

Those are the values I got from the photo and are relative to 0dB.

Using the values I obtained from the photo and calculating the dB relative to the fundamental, the program you linked to returned a %THD of 0.02983268246740213. That looks much more realistic. Interesting considering the Tripath app note shows that their EVB has less than 0.02% THD at 1W (no load specified).

Do you (or anyone else) know what I did wrong in my calculations?

motherone · 2005-04-08 10:44 pm

Don't forget that the output caps on the SI are optimized for a 6 ohm load rather than an 8 ohm load. I would guess that changing the values for an 8 ohm load might improve your results slightly.

Then again, I have no idea about THD testing or anything else on what you're doing here.. Just taking a stab in the dark

jackinnj · 2005-04-08 11:46 pm

BWRX said:
http://www.personal.psu.edu/users/b/w/bww129/pics/amp/1kHzTHDtest.mov

I want to use the video as a screen saver.

Pano · 2005-04-09 12:30 am

Nice work, BWRX.
What are you using for spectrum analysis? How far up can it go?

The reason I ask is that it would be nice to find a way to look at the results of modifying the output filtering. Having a look up to ~3Mhz would should show how well the filters are working.

It would also be nice to to look at the 100Khz to 1Mhz range to see what is going on with the "spread spectrum" switching. Unfortunately, this rage above what most audio analyzer do, and below the radio spectrum analyzers.

Can you measure in that range? Sure would be nice to see how much ultrasonic power is getting through the filters. It could help filter design to know what frequencies are present, and what can be done about them.

Just for fun, hook up your o'scope in XY mode across one channel output and look at the phase. It sure goes through some wild gyrations over the audio band.

My measurements and calculations showed that the amp was capable of about 5W RMS at the unset of "ringing"*. Once the amp starts to really clip, power draw goes way up, as you note, and the waveform tops flatten out completely. That's why the chip draws so much more power.

Thanks again for the cool analysis. With the right tools and the right guys, this amp will have no secrets left!

*digging around in the Tripath docs, what appears as "ringing" on the tops of waveforms is a "feature" of the amp. It's the Tripath version of soft clipping. Supposed to be inaudible. Hmm...

BWRX · 2005-04-09 1:24 am

panomaniac said:
Nice work, BWRX.
What are you using for spectrum analysis? How far up can it go?

The reason I ask is that it would be nice to find a way to look at the results of modifying the output filtering. Having a look up to ~3Mhz would should show how well the filters are working.

It would also be nice to to look at the 100Khz to 1Mhz range to see what is going on with the "spread spectrum" switching. Unfortunately, this rage above what most audio analyzer do, and below the radio spectrum analyzers.

Can you measure in that range? Sure would be nice to see how much ultrasonic power is getting through the filters. It could help filter design to know what frequencies are present, and what can be done about them.

I'm using a Hewlett Packard 3561A dynamic signal analyzer. Unfortunately, it can only go up to 100kHz. It would be nice to see what's going on up there, but I doubt the university has any that can handle that frequency range.

panomaniac said:
My measurements and calculations showed that the amp was capable of about 5W RMS at the unset of "ringing"*. Once the amp starts to really clip, power draw goes way up, as you note, and the waveform tops flatten out completely. That's why the chip draws so much more power.

Thanks again for the cool analysis. With the right tools and the right guys, this amp will have no secrets left!

*digging around in the Tripath docs, what appears as "ringing" on the tops of waveforms is a "feature" of the amp. It's the Tripath version of soft clipping. Supposed to be inaudible. Hmm...

Yessir, the clipping could easily drive the chip into thermal protection mode if allowed to play for long enough, especially without any heatsink. I have to take their inaudible "ringing" claim with a grain of salt, but I can't hear anything above 17kHz anyway so as long it's just that inaudible "ringing" on the waveform peaks and doesn't distort the audible waveform then it won't bother me much

My speakers are ~90dB so I never have to turn it up that loud anyway.

And thank you for sharing your modding experiences as well. That's what makes DIY hobbies so much fun.

BWRX · 2005-04-09 2:03 am

i just took a few photos of the output up to 100kHz with a 1Vrms sine wave input. i will do intermodulation distortion testing but i can't find any appropriate resistors at the moment and i don't have enough cables to hook up two signal generators and the signal analyzer. any other suggestions/requests for what frequencies i should test at other than 1kHz?

i also wanted to let everyone know that the LT1083 is an excellent low dropout regulator. today it was tested for load transient response and line transient response and the output voltage deviation for both tests was about twice as good (half as much deviation) as the dc bench supplies available in the lab! certainly more than adequate power for a single SI board.

jackinnj · 2005-04-09 4:00 pm

BWRX said:
I'm using a Hewlett Packard 3561A dynamic signal analyzer. Unfortunately, it can only go up to 100kHz. It would be nice to see what's going on up there, but I doubt the university has any that can handle that frequency range.

If you're at Penn State I am sure you can find an HP3585A there -- 20Hz to 40MHz -- many of the rf analyzers start at 9kHz.

the inaudible (perhaps) 20kHz to 200kHz region is important for examining amplifier stability., at least in analog land.

you might also find that using an analog 1kHz "notch filter" ahead of the analyzer input will allow you more precision in your measurements.

Pano · 2005-04-09 11:34 pm

jackinnj said:
If you're at Penn State I am sure you can find an HP3585A there -- 20Hz to 40MHz -- many of the rf analyzers start at 9kHz.

Now that's cool! Wish I could get my hands on one of those. To see what frequencies and harmonics are generated by the switching would be very useful. 20HZ to 40MHz, perfect!

Sure would help with output filter design. Should help PSU design, too.

BWRX · 2005-04-10 12:41 am

jackinnj said:
If you're at Penn State I am sure you can find an HP3585A there -- 20Hz to 40MHz -- many of the rf analyzers start at 9kHz.

the inaudible (perhaps) 20kHz to 200kHz region is important for examining amplifier stability., at least in analog land.

you might also find that using an analog 1kHz "notch filter" ahead of the analyzer input will allow you more precision in your measurements.

I've been doing my testing in the EE lab and I haven't seen any other signal analyzers around. I just did a quick facility search and found that I can get access to an Agilent E4403B Spectrum Analyzer that has a range of 9kHz to 3GHz! So what would you guys like me to test? I have two stock SI's at my disposal, one with an old board and one with the newer revision. I can use the bench supplies for 12V power, the signal generators for sine wave input, my 20W 8ohm resistors for the load, and my camera for screen capture. The only thing I don't have is a lot of time before I graduate.

Jack, that's a good tip about using the analog notch filter, but I don't really have enough time or know-how (just basic filter design) to build a good one.

In case anyone's interested, here's another image sequence of the output spectrum of my modified amp and psu with a 1Vrms 1kHz sine wave input, an 8ohm load, and a 12V supply voltage. Pay attention to the scale because it does change to keep the harmonics on the screen. You can see that there are practically no harmonics until the output voltage gets to around 7Vrms and then they increase quite a bit with the even harmonics having relatively low voltages and the odd harmonics having relatively high voltages. Solving for THD at 8W (8.017Vrms) into 8ohms from the photo, I come up with 0.0224% with slightly generous harmonic voltage estimates! That's pretty good compared to SI's claim of 0.1% THD+N at 6W into 8ohms.

1Vrms 1kHz sine wave testing

jackinnj · 2005-04-10 3:30 pm

BWRX said:
Jack, that's a good tip about using the analog notch filter, but I don't really have enough time or know-how (just basic filter design) to build a good one.

Texas Instruments has a great PDF on filter design -- print this out and save it! It includes a notch filter which I used to notch out 60Hz -- you can scale it to whatever frequency.

http://focus.ti.com/lit/an/sloa093/sloa093.pdf

Pano · 2005-04-11 11:09 am

Hey BWRX.
If you get a chance to test, it would be nice to see any or all of the following:

10KHz-1MHz with no signal.
Same with different sine waves (maybe 100, 1K, 5K) at about 2 watts RMS. That should show what frequencies are being used for switching, and how much is getting thru the output filters.

A look at the 50KHZ-2Mhz range, same signals as above. Just to have a closer look at the switching frequencies.

Try the same tests at 8 and 4 ohms to see if it makes a difference to the output filters.

I know this is a lot to ask, but see which, if any, you can do. It will help output filter design a whole lot to know that needs to be filtered out, and how well the stock filters are doing.

Thanks for your good work!

BWRX · 2005-04-11 2:14 pm

I'll see what I can do. I'm heading off to the lab right now to do more in depth measurements of the THD and try some IM distortion measurements. Then I'll go to the other lab and see about the switching frequencies on the output.

Panelhead · 2005-04-11 2:29 pm

Thanks

BWRX said:
I'll see what I can do. I'm heading off to the lab right now to do more in depth measurements of the THD and try some IM distortion measurements. Then I'll go to the other lab and see about the switching frequencies on the output.

Thanks, wish I still have access to Good scopes and other test equipment.
I am interested to see if my high speaker impedance is causing problems. They are about 20 ohms average.

George

BWRX · 2005-04-11 2:49 pm

I can try those tests with my two 20W 8ohm resistors in series. I hadn't even thought about using higher impedance loads for the guys that use their headphones with these amps, but I don't really want to spend all my time in the lab

Check back later tonight for results.

Pano · 2005-04-11 9:58 pm

BWRX said:
Check back later tonight for results.

This is way cool, Thanks!

BWRX · 2005-04-12 2:57 am

sorry to say, but i wasn't able to get to the other lab in time to check out the switching frequencies. i had no idea you can't get access after 6pm... anyways, i was able to do intermodulation distortion testing and some better THD testing. these plots are much nicer because i figured out how to capture the analyzer screen on the computer!

IMD 19kHz, 20kHz 1:1
stock 1W @ 4ohm: 0.167% IHF-IM