Ai yai yai, what a time to be in this hobby… The F5m, the M2OPS, the Pearl 3, Lineup’s Cello One, Tombo’s low powered beauty, Minek’s myriad mosfet amps… I can’t decide what to build next. 
I built an ACAmini for my dad last year and he loves it. I may give him the F5m as an upgrade.
I built an ACAmini for my dad last year and he loves it. I may give him the F5m as an upgrade.
The 22k resistor along with the gate capacitance of the jfets is the input low pass filter.
Makes for a nice square wave.
Makes for a nice square wave.
A lot of open-loop gain is required get low distortion levels.
The JFET stage gain = gm(JFET)/Idss*(output FET gain to rail voltage).
Thus choose the JFET to maximize the ratio gm(JFET)/Idss.
How should I read your FFT plot ?
Fundamental is at -25dB.
So when normalised to the 1kHz fundamental, H2 is at -75dBr, and H3 at -85dBr.
Is that correct ?
Of course now the P3 is no longer possible.
But then large H2 is fashion of the day, right ? 🤓
Patrick
My FFT plot was normalized to absolute ADC levels. The fundamental at -25dB corresponds to 2.82Vrms output followed by a 10:1 attenuator.
The large H2 is not particularly noticeable.
A friend with good ears did A/B listening comparison of my F5m against my ZD25 approximation to a Pass Labs XA25, with volume levels equalized. I did not tell my friend which amp was A and which was B. After a fairly lengthy session my friend could decided he could not tell the difference between them. Neither could I.
My speakers are the Jeff Bagby Kairos with the Kairos Continuum Three Way Woofer module. Their sensitivity is about 86dB, with a nominal impedance of about 5-6 Ohms.
My listening room is a fairly large L shape with a volume of about 275 cubic meters (9700 cu.ft.).
My speakers are the Jeff Bagby Kairos with the Kairos Continuum Three Way Woofer module. Their sensitivity is about 86dB, with a nominal impedance of about 5-6 Ohms.
My listening room is a fairly large L shape with a volume of about 275 cubic meters (9700 cu.ft.).
Watched Nelson's presentation from BAF2023. I have a question: how F5m can have mostly H2 (@lower power level) when it is a scaled version of F5 that has mostly H3?
Just guessing, but H3 dominant might suggest that H2 has been reduced by very careful cancellation. So perhaps if one worries less about cancelling 2nd order, then H3 no longer dominates.
Different MOSFETs used in the original F5.
But if P3 were there, you could have cancelled out H2.
Patrick
Last edited:
I performed some more noise measurements to compare the switching PS noise to that of the linear PS.
The first image is from output for the left channel with the linear supply. The noise at 60Hz, 120Hz and 240Hz is visible. The "noise" at the above 35kHz is from the noise shaping of the ADC.
The second image is from output for the right channel with the Mean Well SMPS. The -124dB 60Hz peak is instrumentation noise. The narrow -124dB 4kHz peak is probably from the SMPS.
Things get interesting between 50kHz and 60kHz, which is shown in more detail in the third image. There is a main peak at around 55kHhz with sidebands at approximately 55kHz +/- 4kHz.
The first image is from output for the left channel with the linear supply. The noise at 60Hz, 120Hz and 240Hz is visible. The "noise" at the above 35kHz is from the noise shaping of the ADC.
The second image is from output for the right channel with the Mean Well SMPS. The -124dB 60Hz peak is instrumentation noise. The narrow -124dB 4kHz peak is probably from the SMPS.
Things get interesting between 50kHz and 60kHz, which is shown in more detail in the third image. There is a main peak at around 55kHhz with sidebands at approximately 55kHz +/- 4kHz.
Thank you for sharing. Do you think RLC filter from the store will improve that 55khz picture? BTW not sure it is all audible.
Any time that SMPS are used there are two sources of noise that may be of concern.
The first is leakage of the AC line frequency and its double; so 60/120 Hz or 50/100 Hz. This can be audible if it is not filtered sufficiently.
The second is the internal oscillator, typically from 50 kHz to 70kHz, depending on the SMPS model. This is generated by the SMPS itself, and is usually not audible by itself unless there is a ‘beat’ frequency with another strong signal that is close in frequency. When a second SMPS is used to create a bipolar power supply, the two units will almost certainly have unequal internal oscillators, and the difference between the two will make a beat frequency. The beat frequency could be anything from 1Hz to 100Hz (very roughly). So this needs to be filtered to keep it from being audible. Other sources of signals that may beat against the internal oscillator are also possible.
The first is leakage of the AC line frequency and its double; so 60/120 Hz or 50/100 Hz. This can be audible if it is not filtered sufficiently.
The second is the internal oscillator, typically from 50 kHz to 70kHz, depending on the SMPS model. This is generated by the SMPS itself, and is usually not audible by itself unless there is a ‘beat’ frequency with another strong signal that is close in frequency. When a second SMPS is used to create a bipolar power supply, the two units will almost certainly have unequal internal oscillators, and the difference between the two will make a beat frequency. The beat frequency could be anything from 1Hz to 100Hz (very roughly). So this needs to be filtered to keep it from being audible. Other sources of signals that may beat against the internal oscillator are also possible.
^ Hard to tell what is happening with this pair of SMPS. There are certainly sidebands around 55kHz, as mentioned. The beat frequency may be low enough (a few Hz) that it is being filtered out.
PKI: There are two 24V SMPS modules. So there could be a beat frequency.
I do not understand the source for the 4kHz (approx) spike that shows in the 2nd image. It is very stable in frequency, unlike the 55kHz neighborhood.
Why should one worry about noise at these low levels?
I do not understand the source for the 4kHz (approx) spike that shows in the 2nd image. It is very stable in frequency, unlike the 55kHz neighborhood.
Why should one worry about noise at these low levels?
Last edited:
Another comparison of the SMPS vs. linear supply.
Both channels driving a 6 Ohm load, 1kHz, 25 Watts.
The first image is the spectrum of the right channel with the SMPS.
The second image is the left channel with the linear supply.
The 120Hz intermodulation (sidebands) in the second image are at such a low level that they are probably of no concern.
Both channels driving a 6 Ohm load, 1kHz, 25 Watts.
The first image is the spectrum of the right channel with the SMPS.
The second image is the left channel with the linear supply.
The 120Hz intermodulation (sidebands) in the second image are at such a low level that they are probably of no concern.
Interesting study! I think anyone would be happy with either of those setups. You are getting very clean results, according to my layperson eyes. 
That's what I'd hope for. Trading off 120Hz for inaudible ultrasonic noise. Not sure about the 4kHz. It could be an alias. If the supplies and loads are identical, I would agree with Tungsten that any beat should be in the LF, but even with a fixed PWM architecture (and even worse with PFM, pulse-skipping, etc.), maybe unit-to-unit variation with slight load differences could produce beats higher into the audible band.
That noise floor looks pretty great, though! Congrats.
That noise floor looks pretty great, though! Congrats.