[Lf] [Fwd: LF: DDS Sources in receivers]

[Andre' Kesteloot]

Klaus von der Heide wrote:

> Andy, G4JNT wrote:
>
> > Unfortunately, the output of the numerical controlled oscillator has to
> > be turned into a real signal, and as it represents a quantised series of
> > angles, a sine ROM must be used to generate a sinewave.  Any other
> > mapping would generate an unpredictable set of non harmonically related
> > spurii.
>
> DDS means direct digital synthesis, whatever you choose to generate
> the function (sine in most cases) digitally. Normally a phase
> accumulator pa of 32 bits is incremented by a constant phase step at
> the sample rate. The sine phase then is  p = pa*2*pi/2^32.
> sin(p) can be calculated from this phase at any accuracy you want,
> by rational Chebyshev-approximation for example as it is done on a
> normal computer. Spurs occur by imprecise sine implementation. The
> usual way is to use a sine-ROM that is addressed by the first 8 bits
> of pa. If the contents of this ROM is directly used as output, that's
> wrong when the last 24 bits of pa do not vanish. In the case of
> 8 address bits, i.e. a ROM of 256 sine values, the amplitude-error is
> up to 2% which is worse than a quantization by only 6 bits. That is
> the error, you mentioned: The phase, that originally is quantized in
> 32 bits is again quantized into 8 bits, which produces an amplitude
> error of up to 2%. Yes, you are right: so it is, if you do so.
> If on the other hand the sine is computed at the correct phase of
> 32 bits, this error does not occur. In my last mail I showed the
> Taylor-series which is nice for sine generation. Using that series
> with the same sine table of 256 values, the amplitude error drops
> down to 0.000000005, which is more than you need.
>
> Because of the importance of the DDS I made two impressive programs
> for demonstration in my lectures:  Both are DDS-realizations with
> two parameters:  (1) Length of the sine table; (2) number of bits
> to represent the sine values in the table.
> The first program is in MATLAB. It shows the spectrum with all spurs,
> the second is an audio demonstration programmed on the DSP56002EVM.
> Using simply a small sine table without interpolation sounds like
> many birds in the background. Even if a sine table of only 16 entries
> is used (4 address bits) no spurs can be heard, when the Taylor-
> series works (then spurs < -60 dB).
>
> >         A DDS output unfiltered is not a square wave.  The quality /
> > resolution in this ROM sets the spurious output level and for the AD9850
> > this is sufficient to meet the specified 60dBc output purity spec.
> > To generate a squarewave, you need to take the filterd sine output and
> > apply it to a comparator.  The AD9850 chip includes such a comparator,
> > capable operating up to beyond 40MHz
>
> That is a very useful feature of this chip. I use two AD9850 in this
> digital mode combined by digital logic to generate square waves with
> edges that can be adjusted in extreme fine steps by program.
>
> > I am about to make a start on a custom LF receiver based on a 455kHz IF
> > and surplus 300Hz mechanical filter, all locked to a master reference
> > oscillator.  A DDS will be used as the Local Oscillator (properly
> > filtered of course !) and possibly another one for the BFO although this
> > is not so essential if a constant output tone of 1kHz is deemed
> > satisfactory
>
> As the previous discussion showed, a filtering of the DDS-signal does
> not help against the many spurs very near to the sine frequency. It
> is better to downconvert to AF by say 128 kHz and to do the tuning
> and filtering digitally by sound card or DSP. Or sample the 136 kHz
> directly and send the result via the USB-Port to the PC. The latter
> really would be a revolution in amateur radio.
>
> But, if you plan your original structure, you possibly should think
> about one more AD9850, that can be programmed to generate any sample
> clock synchronously to the other frequencies.
>
> Now I have to leave this reflector for six weeks of holiday.
> Have nice discussions and luck on LF!
>
> 73 de Klaus, DJ5HG