SMPS vs LPS effect on Garage1217 Project Sunrise III

Discussion in 'Headphone Amplifier Measurements' started by atomicbob, Sep 15, 2017.

  1. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Sep 27, 2015
    Likes Received:
    Trophy Points:
    On planet
    Several points:
    1. do not confuse time domain linear scale with frequency domain logarithmic scale.
    2. time domain signal is a composite of the frequency domain components - it is necessary to include all the harmonics to arrive at the correct time domain amplitude.
    3. there is a 60 Hz amplitude modulation evident on the SMPS as well as many other harmonics. The ripple is not a simple 500 Hz sine wave. See examples below.
    4. The period of the Jay's audio signal is consistent with that of a 60 Hz sine but is also a complex waveform, not a simple 60 Hz sine. See examples below.
    5. Zoom the U8001A picture to see the 60 Hz noise.

    The difference between the Jay's audio 3 mVpp ripple and the Keysight U8001A 670 uVpp ripple is 13 dB total. The waveforms look considerably different. Keep that in mind while going through the examples below.

    Some examples using a signal generator with sine and arbitrary waveform output. These will be clean, without high frequency noise, for visual clarity.

    60 Hz 200 uVpp ripple - time domain
    01 60 Hz 200 uVpp - 2 mV 5mS div.png
    Looks like a sinusoid.

    60 Hz + multiple odd harmonics 800 uVpp amplitude - time domain
    02 60 Hz + harmonics 800 uVpp - 2 mV 5mS div.png
    Has a period consistent with that for a 60 Hz sinusoid but doesn't look like a 60 Hz sinusoid.
    Note the 12 dB amplitude difference between the two time domain examples.

    60 Hz 200 uVpp ripple - frequency domain
    03 60 Hz 200 uVpp - FFT 50 KHz BW.png

    60 Hz + multiple odd harmonics 800 uVpp amplitude - frequency domain
    04 60 Hz + harmonics 800 uVpp - FFT 50 KHz BW.png
    Note that both frequency domain examples have the same -82 dBu 60 Hz component.

    SMPS 4 mVpp ripple
    ripple 1mVrms 4mVpp 50KHz BW filter 2mV 5mS div - annotated.png
    Note the amplitude modulation and also the thickness of the traces. Amplitude modulation are mains noise. Thickness of traces indicate noise in the spectrum. While the trace does have similarity to a simple sine, this is still a complex waveform which when converted to the frequency domain will have plenty of components other than the obvious one at 500 Hz. Keep in mind the previous example demonstrating a 12 dB difference in time domain amplitude while having the same 60 Hz frequency domain component.
    • Like / Agree Like / Agree x 11
    • List
  2. Forza AudioWorks

    Forza AudioWorks MOT: Forza AudioWorks

    Jun 20, 2017
    Likes Received:
    Trophy Points:
    Nicely explained, thanks!
  3. trl

    trl Acquaintance

    Apr 23, 2017
    Likes Received:
    Trophy Points:
    Iasi, Romania
    A similar effect I've seen on a DAC/headamp I'm testing it these days, just the noise goes through entire audio circuitry till the headphone plug. Basically, lot of DACs have a single PSU to power up (something between 5 to 12V), even if inside there are used higher voltages to get more voltage from the output buffer. Of course, I'm not speaking about high-end devices here, but what matters most is the effect of the PSU ripple & noise on the final sound. Without basic measurements I don't trust my ears. :)

    I'm going to present you, from mains to the 6.3mm output plug:

    230V/12V SMPS, most of the time the +12V rail looks very good for a SMPS:

    ...but sometimes it does that:

    XLSEMI XL4005, +15V DC step-up converter (operating around 300KHz)

    XLSEMI XL6008, -15V DC step-up converter (operating around 400KHz)

    XL4005 -15V AC ripple & noise

    XL6008 +15V AC ripple & noise

    Headphones out noise (no music on input, volum to zero)

    So indeed, if PSU noise gets higher than what can be filtered/rejected by the PSRR/CMRR of the opamps (or discrete circuitry) used, then headphones output could become noisy, even audible with very sensitive cans. Not mentioning that some noise could also get injected into the GND (I've also seen this nasty behaviour on ARM chips). The above noise is located between 300-400 KHz, where DC-DC converters operate, so most of the noise is only measurable and not audible, but there are still enough harmonics that fall into audio spectrum.

    However, after removing both DC-DC buck converters and wiring a dedicated LPSU I was able to fix the ripple, though without using very sensitive cans the noise was not audible or it was negligible.

    Linear PSU with Hammond case: +12V, +/-15V.

    Note: Excuse the 8-bit resolution, but didn't got the time to borrow my friend's Tektronics.
    • Like / Agree Like / Agree x 4
    • List

Share This Page