I would argue "full blast" can be interpreted three different ways:
1) Amp driven to produce the highest SPL the user is comfortable with.
2) Amp driven to peak dissipated power. This happens when a sine wave is produced at half the rated output power.
3) Amp driven to just shy of clipping.
Of these scenarios, 2) is the worst case for dissipated power, hence, the worst case for the thermal design.
My point exactly. In the interest of precision: The limit is by the thermal resistance from the junction to the heat sink (of which the case-sink interface is a sizable part).
I'm not sure where you get the 10-30 dB from. Sound-on-Sound analyzed 4500 tracks and found the average crest factor to be 14 dB. The range was from about 7 dB to 20 dB. The article is available here: SOS Loudness War. I cover the math in the section called The Crest Factor on my thermal design page.
SPiKe is a temperature sensor integrated into the output stage. From the first page of the LM3886 data sheet:
The thermal time constant on-die is very short. This is why the die temperature can shoot up to 150 ºC in an instant even though the package remains at, say, 40 ºC. You can convince yourself of this by measuring the temperature of the LM3886 while performing a sweep of the input voltage (as is done in the THD vs power sweeps).
~Tom
Last edited:
Member
Joined 2006
No direct experience but I often read that AD797 is a good sounding opamp.
Thanks. Will have a look on page 19 🙂
Last edited:
So these would be the usual sighted, uncontrolled tests which make no attempt to stay within auditory memory, involve no objective measurement, and are unrepeatable due to lack of documentation? If you find anything with enough rigor meaningful conclusions can be drawn please do link it. Since the testing is likely of different topologies and higher output impedances it's probably not directly applicable to this thread's scope. But it'd at least be an increment of usable data.
The ABX I mentioned in post 134 is from a composite topology. I suppose I should dig up the write up I did on it years ago. I think it was in the solid state forum...
The ABX I mentioned in post 134 is from a composite topology. I suppose I should dig up the write up I did on it years ago. I think it was in the solid state forum...
IM not so HO, the 'good sound' of Tom's amp is MOSTLY due to the care, experience and testing he did for the PCB layout and matching all of it to the parts he used. The circuit & parts are only a small part at this level of performance.
If you use a different OPA, expect to spend at least as much time & trouble as Tom to get it right.
But of course a PCB doesn't look "hand carved from solid Unobtainium by virgins" so can't command the $zillion price that Mains cables or speaker cable spike supports do. 😱
Of course if Virgins & Unobtainium is your thing, I can sell you some labels made by same. Sticking these to your 3886 will result in much improved clarity & definition. 🙂
HOW DO I KNOW ALL THIS?
Cos in my previous life, I was a true Double Blind Listening Test bla bla guru. Absolute Listening Tests-Further Progress
That sums it up rather nicely. Thanks.
Oh, no! Not a SCIENTIST!!! Must go hide now... 🙂
On a more serious note, The University of Surrey and B&O have a nice AES paper on the some various cognitive biases involved in listening tests. The paper is available for free here: On Some Biases Encountered in Modern Audio Quality Listening Tests—A Review. Table 1 on page 3 of the .pdf sums up the biases they looked at in that paper. Those who dare to look at this paper will notice that even the best designed listening tests have significant issues, which results in noisy data and/or systematic shifts in the perceived sound quality scores.
For those further interested in cognitive biases, I suggest Matlin, "Cognition". For easier reads, Dan Gilbert's "Stumbling on Happiness" and Dan Ariely, "Predictably Irrational" are good reads as well. Of course, all these biases are old news to the marketing folks. They've long exploited our various biases and irrationalities to make us buy more products...
~Tom
Last edited:
I did some messing around with the Modulus-86 tonight to address the questions about performance at 4 Ω load.
I'm still using the isolated package and am getting solid performance up to above 50 W into 4 Ω. The amp clips at 65 W but the performance starts degrading within 1 V from the max output voltage with a 4 Ω load. I'd rate the amp as a 50+ W amp into 4 Ω at this point.
I was using a linear, unregulated supply so it's entirely possible that a larger reservoir cap would yield better performance. I was using 2x 22000 uF.
I have no trouble getting to the specified 65-68 W (0.1 % THD) without hitting the thermal limiter, as long as the heat sink is big enough. I'm using a (large!) heat sink with a thermal resistance of 0.4 ºK/W. The heat sink temperature stabilized at 43 ºC at the peak power dissipation (at Pout = 40 W into 4 Ω).
~Tom
I'm still using the isolated package and am getting solid performance up to above 50 W into 4 Ω. The amp clips at 65 W but the performance starts degrading within 1 V from the max output voltage with a 4 Ω load. I'd rate the amp as a 50+ W amp into 4 Ω at this point.
I was using a linear, unregulated supply so it's entirely possible that a larger reservoir cap would yield better performance. I was using 2x 22000 uF.
I have no trouble getting to the specified 65-68 W (0.1 % THD) without hitting the thermal limiter, as long as the heat sink is big enough. I'm using a (large!) heat sink with a thermal resistance of 0.4 ºK/W. The heat sink temperature stabilized at 43 ºC at the peak power dissipation (at Pout = 40 W into 4 Ω).
~Tom
The more this thread goes on the more doubts I have...
That's a HUGE bank for what I would call a low power amp... especially one that opened with claims of excellent PSR. Was that with one board???
I continue to recommend +-20mF for almost any size of wideband Power Amplifier.
20mF allows a 80ms or so Input filter and that allows exemplary low bass performance down to 20Hz.
If you want to drive difference impedance speakers I recommend you scale the required capacitance inversely with the impedance.
i.e. 16ohms & +-10mF, 4ohms & +-40mF
20mF allows a 80ms or so Input filter and that allows exemplary low bass performance down to 20Hz.
If you want to drive difference impedance speakers I recommend you scale the required capacitance inversely with the impedance.
i.e. 16ohms & +-10mF, 4ohms & +-40mF
Pretty sure Tom was referring to the swing limit imposed by ripple causing clipping, not the "noise" limit associated with PSRR. If so, I can't see why this would cause doubts; it's a class AB amplifier, not a buck/boost converter.
Exactly. On the FFT, it's pretty clear that the 120 Hz ripple starts limiting the output swing once the output signal nears the rails. This is hardly surprising.
I'm not sure what there is to be nervous about. No amplifier that I know of can provide an output voltage that goes beyond the supply rails. If the supply rails have ripple on them, the peak output voltage that can be provided will be time-dependent. At the crest of the ripple, the amp can provide a slightly higher output voltage than it can at the valleys of the ripple voltage. This applies to any amplifier.
The only thing that I do, that is different from other amplifier designers, is that I put all the cards on the table. My design is the best documented design out there. You know exactly what you get when you buy it. If this makes you nervous, then I can't help you... 🙂
~Tom
If you plan to use the amplifier to drive a continuous sine wave at the max output power, those are good starting points. For music signals, you could probably reduce the capacitance by a factor of 2~4x, as the crest factor of typical music is considerably lower than that of a sine wave.
I don't know why you would reduce the capacitance, though... A 22000 uF (aka 22 mF) 50 V cap of a well-known brand (Panasonic, CDE, etc.) is $3.50 at QTY=1 from an wide array of distributors, so why skimp?
~Tom
I test my amplifier builds to maximum power and a little beyond.
But all are designed for reproducing audio and music.
I still recommend +-20mF for all my 8ohms capable music & audio reproducing amplifiers.
Changing the load impedance does require a different reservoir capacitance, even if only for the changed current demand.
Hello Tom!
I can confirm your design document is very clear and demonstrate how much energy you put on the details, including in hte choice of the component as per the BOM.
Have you done any tests with the SMPS300RE ...? I m about to buy it but still hesitating with linear apporach, you dont seem to recomend snubbers anywhere on the linear PSU or at the input of the board, right ?
I can confirm your design document is very clear and demonstrate how much energy you put on the details, including in hte choice of the component as per the BOM.
Have you done any tests with the SMPS300RE ...? I m about to buy it but still hesitating with linear apporach, you dont seem to recomend snubbers anywhere on the linear PSU or at the input of the board, right ?
The PSRR of his design is so outstanding that the amp is about as indifferent to power delivery as any I've ever heard of. I think Tom said he couldn't measure any difference in distortion between his lab testing supply and a SMPS. The discussion about the capacitance already took place above. Tom does use snubbers where it matters most; near the outputs where they keep things stable.
Thank you.
I have. My initial conclusion was that the SMPS300RE degrades the performance of the MOD86 slightly in a wide-band THD+N measurement. This is likely because the switching frequency of the SMPS300RE increases the noise floor locally around the switching frequency. Give me a couple of days to get more results and investigate a little further.
One note on the SMPS300RE: The supply is rated for 300 W assuming a crest factor of 2.5. I missed the latter part and ran it at about 230 W RMS for a bit to test it. After about five minutes, the supply died. One of the diodes on the secondary had overheated and fried. I have it all fixed up and have tested it at 115 W. It performs just fine there.
You can use snubbers if you want to. I don't think they matter, honestly. In a previous project I looked extensively at snubbers across the rectifier diodes as well. I couldn't measure any difference with/without the snubbers when using a fast, soft-recovery diode bridge.
I compared a lab supply (Agilent E3632A) against a linear, unregulated supply (toroid + rectifier + 2x22000 uF) and saw no difference in performance between the two. Yep. That speaks volumes to the supply immunity of the Modulus-86.
~Tom
Last edited:
I paid $85 including shipping to the US for the SMPS300RE that I have. I went with the ±30 V version and will turn it down to ±28 V. The ±24 V version would be a good candidate as well.
The lead time and transit time are the only drawbacks that I can see right now. The performance of the supply could probably be improved by adding a power-factor correction circuit in front, but it seems to work well enough.
~Tom
The lead time and transit time are the only drawbacks that I can see right now. The performance of the supply could probably be improved by adding a power-factor correction circuit in front, but it seems to work well enough.
~Tom
Connex have them for $65, but not checked shipping. Lead time is not a worry as this will be a staged build. That barely buys a decent toroid.
I went directly to Connex. $65 + $20 in EMS shipping. Pretty fair. With an Antek power transformer, I can build a power supply at a lower cost, but the savings in weight and bulk of the SMPS are nice.
~Tom
~Tom