Review and Measurements of Purifi 1ET400A Amplifier

[Hayabusa]

I got a question regarding this results that I will appreciate some feedback:

The High frequency peak is still only -20db from the main music that sits on 0 ?
Isn't that basically means that we spend 0.1W for each Audio level we product - and that is literally being dissipated on the tweeter ?

On the Starkrimson-Ultra - at least in the graphs published, it shows that the Audio signal and the 750KHZ are of the same height.
So - does that means that it basically the tweeter is going to be driven (just power power dissipation ) x2 harder than the audio signal calls for ?

For some reason I thought that the filter is supposed to be much stronger no ?

Thanks

Its not clear from the measurements at what level the reference 1KHz tone is. (@amirm: may be you can tell?)
I assume its much lower than full power.
The switching harmonics are to some extend input level independent.
For example if the 1 Khz was at 1watt the harmonics are at 0.01 watt only and will not be significantly higher at higher output levels.

 

[RandomEar]

I got a question regarding this results that I will appreciate some feedback:

The High frequency peak is still only -20db from the main music that sits on 0 ?
Isn't that basically means that we spend 0.1W for each Audio level we product - and that is literally being dissipated on the tweeter ?

On the Starkrimson-Ultra - at least in the graphs published, it shows that the Audio signal and the 750KHZ are of the same height.
So - does that means that it basically the tweeter is going to be driven (just power power dissipation ) x2 harder than the audio signal calls for ?

For some reason I thought that the filter is supposed to be much stronger no ?

Thanks

Using a first order approximation, the power is proportional to the surface area below the graph. If you play just a 1 kHz test tone at 0 dB, that switching tone at -20 dB would carry roughly 10% of that power. But if you play music, there's much much more than a single 1 kHz signal being generated. The surface area below the music signal would therefore be significantly larger, too. If you then compare that larger area to the switching tone, the power ratio will mostly trend towards zero.

That being said, a better suppression of the switching noise would definitely feel better, even if it doesn't matter from an engineering point of view.

 

[Hayabusa]

a 1 kHz test tone at 0 dB, that switching tone at -20 dB would carry roughly 10% of that power

not 1%?

 

[RandomEar]

not 1%?

Yes, thanks.

 

[ADanalog]

I am confused on the unit of [dBrA], is that an voltage amplitude or power ?

In any case - I would think that the correct test method would be to run it on the 32 tone test, and insect the high tone element

what do you think?

 

[KL....]

I am confused on the unit of [dBrA], is that an voltage amplitude or power ?

In any case - I would think that the correct test method would be to run it on the 32 tone test, and insect the high tone element

what do you think?

Thank you, @ADanalog, Yes, they like to confuse here, don't they, which is a little sad, isn't it?
dBrA is a specialized unit of measurement representing decibels relative to an analog reference level (usually 0dB), with A-weighting/filter applied. It is commonly used in professional audio analysis (such as Audio Precision software) to quantify signal-to-noise ratios (SNR), dynamic range, and frequency response, often in relation to a 0dBFS (full scale) digital reference.

• dBrA vs. dBA.... while both use A-weighting, dBA usually refers to sound pressure levels (SPL) in the air, whereas dBrA usually refers to signal levels in electrical circuits.
• dBrA vs. dBFS.... dBrA indicates a relative measurement, while dBFS (Decibels relative to Full Scale) is the standard for measuring digital audio levels where 0 dBFS is the maximum possible signal.

 

[boXem]

In any case - I would think that the correct test method would be to run it on the 32 tone test, and insect the high tone element

In the specific case of out of band noise, what would be shown that is not seen with single tone?

 

[ADanalog]

Thanks KL ,

So the if the measured signal is in dBrA, I am assuming that it should be referenced to a the main tone amplitude? , will it follow [20 * Log (V_measure/V_Ref)], and that residual signal is ~10% of the audio signal (I hope I interpreted it correctly )
I need to get home a test my amp on a 4oh, load and see the actual residual signal at 750KHz, I remember that I tested it .

Classic concept of an LC filter which is what the "classic low pass" indicate that the signal should have decayed by 1/100 as it is a 2nd order filter, with the assumption that the amplifier BW is the LC filter corner-Freq, but that is assuming that the modulation changing its amplitude with the signal being amplified.

On this thought I found the test in the below link - they have two tests : one showing -40db fall between the 1KHZ and the 750KHz, but in the square wave - the residual ~2Vpp out of the square of ~9Vpp.

Why would it be like that ?

SoundStageNetwork.com | SoundStage.com - Orchard Audio Starkrimson Mono Premium Amplifier Measurements

www.soundstagenetwork.com

 

[nagster]

I am confused on the unit of [dBrA], is that an voltage amplitude or power ?

In any case - I would think that the correct test method would be to run it on the 32 tone test, and insect the high tone element

what do you think?

About dBrA in Audio Precision software.

 

[rdenney]

I am confused on the unit of [dBrA], is that an voltage amplitude or power ?

In any case - I would think that the correct test method would be to run it on the 32 tone test, and insect the high tone element

what do you think?

It's voltage amplitude used as a reference for analog inputs. It is settable for various test protocols, per the instructions in the previous post.

The traditional approach for measuring distortion and noise is to feed a clean signal of known input voltage (RMS, of course) to the device under test, and then for the signal from the device under test, electrically or digitally notch out the test signal and measure the true RMS voltage of what is left. This will measure both distortion and noise, of course. This is the method used by standalone distortion analyzers such as the HP 8903b. If the harmonic frequencies are all notched, the residual voltage will be non-harmonic noise, though the notch filters will start to overlap as the harmonics get closer together, introducing error at high frequencies. I don't know of any traditional analyzers that do that, and usually the plain signal/noise ratio is measured signal magnitude (RMS voltage) divided by the RMS voltage of quiescent noise. Harmonic distortion resonances that don't rise above the noise floor are invisible, but also inconsequential.

The same strategy could be used for a multitone test, with the great improvement that the tones are spaced linearly rather than geometrically as with harmonic overtones, so they don't get closer together at higher frequencies, but I don't think this was easy to do before software-driven analyzers.

Another traditional approach is to use a spectrum analyzer and observe/measure the amplitude of resonant spikes. This is an approximation, because it's not a true RMS measurement of amplitudes away from the test frequency, but it does identify the magnitude of distortion products in a useful way. Noise is stochastic, so the noise floor in a real-time FFT will be chaotic, but it's usually pretty easy to visualize the magnitude of the noise floor with useful precision.

When the signal is digitized and evaluated digitally in modern software, the FFT analysis already provides the details of the shape of the frequency spectrum with sufficient resolution, and it's easy to average the amplitude of noise and distortion products, and even notch out the harmonics numerically.

(I mention traditional practice here only as an explanation. I don't think any standalone distortion analyzers of old had a low enough noise floor and a clean enough reference signal to measure the performance of the best equipment now available, and that certainly includes the Purifi modules. My HP8903b is about as good as any that were made, near as I can tell, and it can measure distortion and noise down to about -100 dB from a 1 or 2V reference signal.)

But it's all voltage ratios.

Rick "learned a lot about this playing with his vintage HP distortion analyzers" Denney

 

[ADanalog]

Thanks Rdenney and Nagster.

This class D measurements are really complex concept.

Without a signal being modulated (at least from TI's lates chips) - the P&N sides will have full swing identical switching signals before the filter - so this actually should generate the most common modes signal, and it looks like TI and Purify using techniques to spread the spectrum (this will just be EMI, as the differential signal will sum to zero on the speakers), the speakers will see zero differential signal.

The signal to be amplified, is causing a differential signal on top of that common mode switching residual signal, and that will have a component of the switching signal too.

The a Constant differential level into the gain stage (square wave) is causing the most "stable" constant PWM , Since the differential signal is the main one that will show the high speed switching - it seems to indicate that the "right stress point" to estimate the amount of switching power dissipating on the tweeter , is to test the residual modulated effect at the lowest frequency with a tone of 20HZ - This will generate the largest residual ~750KHZ .

Does this makes sense ?

Thanks