FL2K replace MT34TL/AS11D buck chip with LDO

historical

I'd like to expand on that: the fl2k is just a DAC without
reconstruction filtering. Like every DAC without a filter, harmonics
mathematically must exist. That's not a bug or something due to
imperfect design – that's just the theory behind DAC; especially: has
nothing to do with the power supply :)

In essence, you can think of a (perfect¹) DAC as something that
produces a series of extremely sharp impulses of the correct amplitude
at sampling rate, with zero in between. So, let's imagine these
amplitudes would form a nice cosine of say 20 MHz if you "connected the
dots" – but you don't do that connecting electrically just yet.

           *   *
       *   |   |   *
+--*---+---|---|---+---*---+---|---|---+---*-- …
*                          *   |   |   *
                               *   *
   ^                                      ^
   |____________ T=1/(20MHz) _____________|

When you think of the spectrum of that, I'd say we both agree that
there's power at 20 MHz, so obviously the PSD must have a peak at 20
MHz. But hey, if this was *actually* a sine of 20 MHz, then there would
be "smooth" connections between the signal impulses (the "*" in my
ASCII drawing). There are not; so, there must also be a signal
component which "suppresses" the 20 MHz sine in between. Whilst there's
fine math theory which lets us derive this directly, I simply imagine
there being 1-0-1-0… wave that I multiply with the "pure" sine to
arrive there – first one of twice the sampling rate, to "zero out" the
middle between amplitude instants, then another one with higher
multiples of the sampling rate and so on. 

As someone dealing with HF, you probably know what happens when I
multiply one sine with a wave of a different frequency: You mix one
tone by the other. And that's exactly what you'll see at the output of
*every* DAC: the spectrum you produced in the baseband, and a
repetition every sampling rate distance. These repetitions are
fittingly called /images/, and you often (and you, especially, in this
case) suppress them with a simple low-pass filter. The effect of that
/reconstruction/ filter is that it actually smooths out these impulses
– it "connects the dots"!

Good thing is that the sampling rate of the fl2k is plenty high enough
to cover your HF band at once – you don't need to rely on this imaging
to get to any portion of that band. In case you haven't seen that, the
fact that imaging exists is even useful: By selecting an image that is
far above the baseband, you can, with a DAC that has ~160 MHz sampling
rate, even generate signals in the 900 MHz GSM band, for example,
without any mixer.

So, what you need is a filter with a cutoff frequency above HF and
below half your sampling rate.

By the way, power supply effects *can* lead to harmonics, but usually
in amplifiers: when your power supply is insufficient for your
amplifier, then that amplifier can't reach high amplitudes as fast as
it should – that's a nonlinearity. Nonlinearities lead to mixing, i.e.
harmonics.

My suspicion is that you meant "spurs", not "harmonics", caused by the
buck converter. You're referring to
http://people.osmocom.org/steve-m/fl2k_slides/osmo-fl2k.html#(17) ,right? 
Well, in fact, these seem to be harmonics, but my best guess it's
they're harmonics of the boost switching clock, mixed with the signal
of interest, not of your signal itself. We typically refer to tones
that aren't harmonics of the signal of interest but are generated
within a device as spurs. Anyway, that's just semantics, in the end.
You want to get rid of them; first, make sure you actually need to do
that - they are present in 1 MHz steps, as it seems on that slide
above, so chances are that if you choose your rates cleverly, you can
avoid seeing them in-band at all.

Before replacing any complete buck converter, I'd try to uppen its
output smoothening: find the large output caps after the inductors, and
parallel/on top solder another 10µF; follow the positive voltage trace
and solder a good 10nF onto where that connects to the fl2k; connect a
very solid wire to ground to the other end. Maybe that's already
sufficient at surpressing things. 

Best regards,
Marcus

-----------------------------------------------------------------------
¹ "real" DACs often work slightly different, e.g. do zero-order hold or
such, but that doesn't actually matter for the principle of images –
what matters is that the "difference" between the "pure" smooth signal
and the DAC's output has frequency components at multiples of the
sampling rate.

On Fri, 2019-02-08 at 07:59 +0100, Sylvain Munaut wrote:
> Hi,
> 
> > Has anyone tried replacing the MT34TL/AS11D buck converter/supply
> > chip with a LDO regulator as Steve Markgraf suggested - my goal is to
> > remove the harmonics, for HF tx.
> 
> Remove the harmonics ? The fl2k will always have harmonics, that's
> just the way it is. Only way to get rid of theses is to post-filter.
> What a clean supply does is remove the spurs.
> 
> Cheers,
> 
>     Sylvain
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 6582 bytes
Desc: not available
URL: <http://lists.osmocom.org/pipermail/osmocom-sdr/attachments/20190208/303d1eaf/attachment.bin>