[Lf] even more on DDS

[Andre' Kesteloot]

Klaus von der Heide wrote:

> From:                   "Graeme Zimmer" <gzimmer at bigpond.com>
> To:                     <rsgb_lf_group at blacksheep.org>
> Subject:                Re: LF: What is the best RX for 136 kHz?
> Date sent:              Tue, 11 Jul 2000 12:54:32 +1000
> Send reply to:          rsgb_lf_group at blacksheep.org
>
> Zim, VK3GJZ wrote:
>
> > The worst spurs are caused by the fact that the DDS is unable to change it's
> > output except at the clock edges.
>
> Definitely no! The sampling theorem in a DDS never is violated. Spurs
> only can be produced by non-linearity. One source is, as Johan wrote:
>
> > > In-band spurious signals are generated due to the limited DAC
> resolution.
>
> DDS means Direct Digital Synthesis. The main source of non-linearity
> is the generation of the sine function itself. Usually a lookup table
> is used plus a cubic interpolation or a Taylor expansion of third
> order. I use the latter in my DDS-implementation on the DSP. It is
> accurate in 23 bits causing no spurs above -135 dB. If the Taylor
> expansion is omitted the lookup table must be very very large to
> avoid spurs (note: it's totally digital, there is no ADC).
> So, if there are spurs I would guess the sine implementation is not
> as accurate as you want. This error vanishes when the interpolation
> is made for time points exactly lying on one of the points of the
> lookup table. That is just your observation:
>
> > At frequencies where the output is an exact sub-multiple of the clock, a
> > very clean signal is obtained.
> > But at all of the intermediate frequency steps, the DDS is actually
> > jittering between the two nearest sub-harmonics.
>
> I guess, then only the nearest table point is chosen for output
> instead of a correct interpolation.
>
> > The spurs caused by the DAC can be reduced by using more bits in the system,
> > but the FM spurs can only be reduced by using a (much) higher clock
> > frequency (there are various other schemes of course).
>
> No, the clock must be more than twice the output sine frequency.
> That's all.
>
> > > Another option is to make the receiver almost totally digital.
> >
> > The trouble here is that any digital RX will still need a digital VCO (NCO
> > or DDS) which will always have the above inherent FM spur problem. If the
> > DDS is in software, it will of course have a much lower clock frequency
> > (instruction step) than the hardware DDS chips.
>
> In a DSP-solution the samples are computed for virtual time steps.
> There is no relation between these time steps and instruction step or
> the processor clock. A conversion from virtual time steps (as on an
> audio CD) to real time is done by the A/D-converter. Therefore only
> the clock of the ADC determines the accuracy of frequencies in
> DSP-solutions, not the DSP-clock.
>
> As I said above, a DDS-routine for a DSP can be programmed to any
> precision. That is the reason why I prefer the almost totally digital
> receiver based on a programmable DSP.
>
> > I would love to be proven wrong in this rather pesimistic analysis...
>
> I hope I achieved that success.
>
> Thanks to Johan and Zim for their comments and to Petr for his root-
> question! Otherwise this clarification would not have been possible.
>
> 73 de Klaus, DJ5HG
-------------- next part --------------
An embedded message was scrubbed...
From: "Klaus von der Heide" <v.d.heide at on-line.de>
Subject: Re: LF: What is the best RX for 136 kHz?
Date: Wed, 12 Jul 2000 09:05:19 +0200
Size: 4512
Url: http://atanasoff.rf.org/pipermail/lf/attachments/20000713/70aea977/attachment.txt