Alesis M1 MKII very simple improvement of bi-amplifier THD by at least -8dB

[EricZenviewStud]

Hi,
Before I reassembled those powered speakers after repairing one of them, I tried some so simple mod
to improve THD of each amp section.

This can be applied to a lot of amps, when they cheap out a bit on capacitors in the power section…

To prove my point, I ‘ll run a similar mod and video on a Rotel 840 BX II, which doesn’t have the capacitors mentioned on
the schematic ! Increasing many times the original capacitance feeding the power section, is needed to obtain a meaningful
THD improvement , here between -8 dB and almost -20dB on the best of the two power sections…

I cheaped out myself putting only 470uF on some of the two channels, depending on which one it was…the better one or the regular one, to give them a name .
I certainly could have pushed somewhat further than the -80 dB THD figure , I obtained overall, but felt that was ok for these amps !

 

[mcdn]

I’m sure the content is lovely but videos without text summaries are not very welcome in these parts.

 

[EricZenviewStud]

Hi M,
Well I thought my text was enough for one to decide wether to click or not, if one had an interest on the subject ?
and it’s true I am not selling anything but being prosélyte to making existing amps / equipment last longer , and thought that thought might interest, maybe wrongly, others .

However trying to write something about a video except that I show how I went to measure , if my changing of capacitors had the effect anticipated or not on diminishing amp THD ? Is that not enough ?

Best,

Eric

 

[cavedriver]

Hi M,
Well I thought my text was enough for one to decide wether to click or not, if one had an interest on the subject ?
and it’s true I am not selling anything but being prosélyte to making existing amps / equipment last longer , and thought that thought might interest, maybe wrongly, others .

However trying to write something about a video except that I show how I went to measure , if my changing of capacitors had the effect anticipated or not on diminishing amp THD ? Is that not enough ?

Best,

Eric

Is this really that common of a weakness of amps, though? Or are you targeting specific amps where it's apparent that the designers built in a weakness? Could a similar modification be made to something like an A/V receiver, where they typically have each amp channel on a small card or built in clusters around the heat sinks? What about a classic, once highly regarded amp that's getting a bit old like an original Adcom GFA-565 or similar - I assume these amps were built without such a weakness?

 

[EricZenviewStud]

Is this really that common of a weakness of amps, though? Or are you targeting specific amps where it's apparent that the designers built in a weakness? Could a similar modification be made to something like an A/V receiver, where they typically have each amp channel on a small card or built in clusters around the heat sinks? What about a classic, once highly regarded amp that's getting a bit old like an original Adcom GFA-565 or similar - I assume these amps were built without such a weakness?

Hi Cavedriver,

Well it’s probably a design choice ie price versus spec and probably few mfg ran tests before hand to see influence of those caps values.

If you look at a NAD 323 Bee you will see that they from origin used 12 000uF caps …and get a THD in the -80 dB ….
The Rotel I worked on had caps of lower value than ( installed by the factory) what was written on the schematic was 10 000 uF, though, I am going to replace them with 20 000uF about three times as much as was installed, and measure what I gain…in terms of THD (next video)

I already molded that Rotel 840 BX 4 , thanks to the reflexion from a British engineer and went from -70 dB to - 80 dB
We will see how much , I can still gain .

To answer the why of your question, I saw a video from someone measuring output power before clipping.
When they changed (doubled) the caps output power before clipping went up by a couple watts…

My take on it, is that those caps act like power tank for the power transistors , when there is not enough “tank” ie instantaneous power, the transistors don’t function as they could…increasing THD diminishing power before clipping.

To alleviate that you increase those caps and induce lower THD with same power and slightly increase power before clipping…

So depending on the design, there is or not some margin of improvement…

My futur tests with 20 000 caps will give some hint what can be expected .

I’ll look after that, at my NAD maybe going for 20 K vers 12 K or more, if possible, to see if that can improve THD further on those amps .

But they are like, on the Rotel surely other sources of improvement on a NAD, just have to find someone who has already done experimentation on it to improve it’s design…

Voila , since those designs, a few designer have truly researched how to improve things, caps for power transistors are the simplest of concepts

The small amps are like the ones on the Alesis, if you look at their schematic, you will the answer looking at the values of the caps feeding the mini power amps…

I tried to answer as best as my current understanding

Cheerios

 

[mcdn]

@EricZenviewStud I've watched the video now, and it took a while before I could understand it by looking closely at the numbers on your analyser screen and checking the speaker schematic and the amp datasheet.

I would summarise it as: "The Alesis M1 mkII uses the LM2876 chip amp, but the manufacturer has not followed the data sheet and has used undersized 100uF bypass capacitors at the LM2876 power supply inputs. Changing these for 470uF or larger parts made a difference to measured THD at very low power outputs. THD+N was unaffected as the measurements were at a very low level, but Eric in the video assumes his Quantasylum analyser is in error. No directly comparable before and after measurements are provided, nor are any measurements at a standard 5W output"

The LM2876 datasheet does show rising THD+N at lower power levels, eyeballing it lets say 0.02% or -74dB at 10mW, whereas -80dB should be achievable at 1KHz/5W

You are also running a fairly low resolution FFT (16k) with a high number of averages (31), which is maybe why visually the noise seems lower than the QA dashboard shows?

I'd suggest redoing the experiment if you have an unaltered speaker to work with, this time running the ASR standard suite of amp tests both before and after, with a 64k FFT wth no averaging. Given the crossover involved it would be fair to choose 1KHz/5W to test the woofer section thoroughly and then 4KHz to test the tweeter section quickly since it's exactly the same apart from the input filter.

Of course if I have this wrong do please correct my understanding, I am reading QA settings off a video after all!

 

[EricZenviewStud]

Hello M,

Thank you for your very complete feedback , analysis and pin pointing what you meant as a written resume , and opening the path for great improvements !

So , the power levels first , I use self made probes for the QA 400 unit , that have selectable -20dB or -40dB attenuations, replacing the more modern analyser versions which have built in attenuators in the unit .

I used the -20dB setting , ie power readings should be increased by 20 dB , thanks for pointing that out !

This is the subject that most likely needs lot more work .

The Video stemmed from an idea, on the simplest of mods to somewhat improve THD, and I ran some tests as quickly as possible to get to see if I could get results and what they were.

My real thought, was about the Rotel 840 BX4 , where I had THD and power readings… although not complete, but much more than on the Alesis which was an afterthought , that my thinking could be extended to it .

So on power :

First according to schematic they are two different amp sections:

High frequency : LM2876T feeding into
25 watts according to manual 4 ohm load ?
Low frequency : LM 3886T feeding into
75 watts according to manual 6 ohm load ?

I haven’t checked any of this load values, but my load were 8 ohms!
quite far away from the supposed values….I did not read parts specs,
my bad , as I wanted to complete this little teaser video in an hour!

(It took me three / four times as much at least !)

So firstly, I will have to get 6 ohm load (or maybe three times 2), to run some further tests !

Also, I should, if I was to be complete, check the actual speakers impedance over frequency …

although I probably won’t have time for that…ie I ll rather spend time to devise a method for comparing almost accurately , their near field frequency response compared to my much better Focal Solo 6 and see if I can devise that way, a simple trick, to make them much closer to true flat response ie much similar to the one I get by ear from Solo6 versus M1…
(To make those Alesis useable for me, or if not sell them .)



According to Manual
LOW FREQUENCY AMPLIFIER
Rated power output: 75 watts, 6Ω load
Distortion: <0.02% / (= -73.9 dB) THD @ 30 watts/8Ω 20Hz-20kHz
Slew rate: 19V/µsec
Signal-to-Noise ratio: >110dB referenced to 60 watts @ 8Ω,
1kHz
HIGH FREQUENCY AMPLIFIER
Rated Power Output: 25 watts, 4Ω load
Distortion: 0.025% / (= -72.041 dB) THD @ rated power
Slew rate: 9V/µsec
Signal-to-Noise ratio: >112dB @ rated output

What I did : I had set my input level to -10 dB, level pot at max level … why ? I did not run precise power measurement , I checked if it warmed up the dummy load and did not start decreasing THD too much , it’s was a super quick hands on empirical approach .Ie not precise !

If I get the time , I’ll take appart again one of the units and will run a new sets of more serious power measurements (and THD).

I did informal testing before doing the video, and I know how these sound ,this how I chose the test frequencies…

Meaning, I felt 1khz would not approximate truly THD response of the low end , one could discuss even not going lower…
They say tweeter response is flat to 2k ! , so 1k is not useful to my hear to run some tests on the low end…this is practice of the field rather than standard testing…
(I consider those drivers and we’re in my mind the most energy is, ie much lower than 1khz)

I confirm not watching too closely some of the calculations reported by QA, and using my own THD approximation ,
because QA is old and somewhat buggy, in some of it’s calculations on screen, I won’t go into all details, but having long software debugging experience, I seem to always step right in were all the bugs and forget fullness are , so it takes me more time to figure if what it is displaying is accurate or not than measuring myself between fundamental and harmonics to get a very precise* idea of where THD lies, (*given the whole picture and the other harmonics..I am not the only one running such approximation on field measurements .)

I ran few times before in loop back my own calculations and confirmed I to be better and quicker off this way…this is a low cost OLD Spectrum analyser, so you as always, have to know your instrument .

Your right on the depth and averaging, but my tries (loop back) , gave me confidence and ease (speed) of use with these setting…64k is slow…
I reset often the averaging, and get a quick feel, what’s seems to be going on .

The power displayed figures are all wrong any how, as the setting of the QA 400 was for 8 ohms load, not accounting for the reality of those amps expectation , 6 and 4 ohms, so in the end I was right on the money to avoid the watts output issue all together in this first try .

Thanks for your time and great help improving, my little experiments ,
this is exactly what I am looking for here…and I am sure there is room for lots of improvements still.

My whole interest is to push the envelope as quickly and easely as possible using gear lying around and trying to innovate a bit with it .

As said , I am not selling anything, rather trying to entice others to run some trials on their own and report back with new better ideas

Cheerios

Eric

 

[markanini]

Thanks! These are my secondary monitors.

 

[EricZenviewStud]

So here is the conclusion !

I devised a simple way without new dummy load resistors see photo , I piggy backed the speakers…

Results are for the best of the two units (that is the one I re-opened)

With 4,64 watts on the low channel (this is what I was running all along but now with proper ohm input in software , still -10 db in, full volume pot ), you can read -95,7 dB THD +N !


On the high channel I was running less than a watt with -10db in , so I raised to 1,78 watts and no impact on THD…you can still read -82,2 THD+N

To make everybody happy I ran a slow 64k ….you can read yourself still -82,4 THD+ N ,
( and I assume that the 6 db lower next closest spur before the rest being -20 dB down don’t really affect the proper calculation… by -3 dB like the -79,3 dB screen figures wants you to believe )

We started around -72 db THD+N, Conclusion : the recipe works !

Eric

PS: one thing I mentioned in the video, but forgot to show was the setup with floating ground the gizmo on the USB port..here is the photo.

 

[AnalogSteph]

These chipamps run relatively lean, so it should not surprise me if they generate a fair bit of switching noise on their rails. The success of increasing rail bypass caps may indicate that power supply routing going back towards the power supply section is a tad on the inductive side. I can't spot anything major on the board layout though.

The stock choice of using 100 nF bypass bypass caps also feels a bit risky, odds are there's an impedance peak somewhere (parallel LC formed by 100 nF + ESL of big caps). Achieving broadband low supply impedance is not entirely trivial. You can probably get away with 2 decade intervals but 3 seems a bit of a stretch. Another parallel 2.2-4.7µ might be a good idea. If film caps sound a bit too big / expensive, check the datasheet for which voltage rating 'lytic gives you the best ESR. My guess is somewhere in the 63 / 100 / 160 V vicinity.

I do like that they seem to have given some thought to tweeter hiss levels, with the final response shaper having overall negative gain and being followed by passive attenuation, plus the absolute minimum stable gain on the power amp:

I don't know whether loading a TL084 or MC33079 with 1k + 1k33 = 2k33 even at the modest levels required for a tweeter is a particularly smart idea though. These NatSemi power amps aren't exactly super low input noise (2 µV(A) is the equivalent of something approaching 20 kOhms), and 1k||1k33 = 571R isn't even a particularly good impedance match for 1k||10k = 909R. Assuming that the tweeter level is fundamentally correct, 1.62k for R10 and 2.15k for R9 would be more like it. It's a pretty crummy divider ratio that requires at least one E48 value unfortunately. That being said, with tolerances on transducers being much less tight than those on resistors, I bet you may well be able to substitute a more sociable combo in this general range. (1.6k/2.2k would only be 0.13 dB off.) I'd tackle that after taking acoustic measurements.

In an active speaker, gilding the lily in the distortion department is not generally required except at low levels. Otherwise the drivers' inherent distortion tends to dominate, potentially by orders of magnitude. This is an advantage over a mere powered speaker with a passive crossover, where amplifier IMD performance is a fair bit more critical.

As a next step, I would focus on acoustic measurements. MMM, tweeter and woofer nearfield windowed, in-room windowed. I have rarely come across a budget monitor that wouldn't benefit from a bit of PEQ.

Plus, it would be interesting to know whether there are any big woofer breakup modes that would benefit from a passive series notch. Those can bring out electrical (motor) distortion or amplifier hiss if left untreated.