At 09:50 PM 9/9/2005, Tayloe Dan-P26412 wrote:
I was kind of excited about the "dithering" idea for a while when I
first heard of it several years ago. The idea behind this is that
since one frequency setting produces a given set of spurs, introducing
"dither", that is hopping over several very close DDS settings such
as four will reduce the spurs. Since the spurs happen only one quarter
the time (in this case, they will be down by 12 db.
Way cool!
However, it then occurred to me that now, even though the spurs are
down by 12 db, there are now four times as many of them. And that
seems a bit counter productive to me.
Lower level spurs, but a lot more of them. Dithering is a tradeoff,
not a solution to the DDS spur problem.
What it does is turn some annoying high level spurs into a flat
distribution. The dithering isn't exactly hopping over different
frequencies. (well.. it technically, is but it's on a sample to sample
basis, so the "modulation index" is effectively very large)
You either add a random number to the phase increment (LSBs) or to the
phase. In either case, the random number is "zero mean" so the frequency
remains constant.
However, you don't get the "not exact multiples of the sine lookup table"
spurs now (or, rather, the energy is spread out)
Another problem is glitch energy in the DAC (which tends to be at multiples
of the clock frequency), especially because it mixes with the DDS output
frequency, and the aliases of the (apparent) mixing products are in band.
Done right, dithering essentially squashes all the spurs down at the
expense of raising the noise floor a bit. Since the spurs are narrow band,
the energy in the spur isn't all that huge. For instance, say you've got a
spur that's -50 dBc. Undithered, the bandwidth of that spur is probably a
Hz at most, so spread it out over 100 times the bandwidth, and you're down
to -70dBc, which is getting close to the quantization noise of a 12 bit D/A.
There's also dithering to reduce the effect of converter nonlinearities,
but that's another is
The challenge is in coming up with a dither that spreads the spurs enough,
while not ruining the close in phase noise of the desired signal (since,
when all is said and done, the dither essentially convolves some sort of
frequency domain response with the impulse like spurs and the desired signal)
- Dan, N7VE
-----Original Message-----
From: Jim Lux [mailto:[EMAIL PROTECTED]
Sent: Friday, September 09, 2005 8:02 PM
To: Robert McGwier; Tayloe Dan-P26412
Cc: FlexRadio@flex-radio.biz
Subject: Re: [Flexradio] DDS discussion
At 06:39 PM 9/9/2005, Robert McGwier wrote:
>Dan:
>
>Thanks for the information.
>
>The AD9958 seems to promise a great improvement over AD9854 which Gerald
>uses in the SDR-1000. If Gerald were not trying to do an all shortwave
>receiver and ham band transceiver, he could easily do the PLL spur
>removal circuit but it is a bit hard to do one that tracks 50 Mhz of
>reference.
It's hard to do in a "direct conversion" architecture. It's much simpler
in a dual conversion architecture (say you converted to an IF of 250 MHz),
if only because the "tuning range" of the DDS LO is a smaller percentage
bandwidth.
> For dedicated ham band units, such as the NC2030 and the Elecraft
>K2, the problem is solved by having no DDS or having a DDS as the
>reference to a PLL circuit and then restricting the tuning range to
>something reasonable. It appears clear to me that this problem of
>spurious emissions from DDS is well on its way to being resolved and
>any future designs for Flex Radio SDR's should take into account that
>almost surely, you will want to replace the DDS section.
Since we're theoretically writing software radio software that's
applicable to a wide variety of hardware (since there's a hardware
abstraction layer), it should be immaterial to the software. HOWEVER,
we're always going to be working with non-ideal DDSes and radios, that's
why the HAL needs to have the ability to give "guidance" to the upper
layers about recommended parameters to set. (this is one of those
horrible breakdowns in the classic encapsulation model)
As to whether a radio should include a "replaceable DDS"... unless it's a
pin for pin drop in, and folks are really ambitious about unsoldering a
200 pin fine pitch IC, I can't imagine it. You've already got one board
that has the analog buffers, the DDS, and the DDS filters on it. I'd just
replace the whole board in the stack. Gerald did a fine job on
partitioning in the original design. If you get too far in the
"replaceable modules" thing, it gets unwieldy, and you wind up with those
6U VME racks of boards so beloved of the signals analysis community. (for
those looking for pain, google for JTRS (Joint Tactical Radio System) and
SCA (Software Communications Architecture).)
BTW, the 9858 may do this, but the not-so-recent hot ticket for NCOs is to
do dithering to spread the spur energy around. There's also a whole raft
of other clever approaches. For these, you're typically looking at
putting the digital part (the phase accumulator, adaptive equalizer, etc.)
in a FPGA and using external DACs. The AD985x approach is a higher
integration, but lower flexibility, approach, because it integrates the
DAC and digital logic.
> The AD engineers are
>really pushing hard. They are also urging great caution about using
>the 9951/9954 in quadrature for multiple reasons. The large spurs are
>at submultiples of the clock
Which is a quite common problem with NCO/DDS architectures. There's a
fascinating treatise on the web (and if I weren't sitting after dinner
with a glass of wine in my hand, I'd remember the URL) with some great
Matlab code that does the whole spur/noise analysis for generalized
NCO/DDSes, including various dither and spur cancellation schemes. It
started as a PhD thesis. All nicely validated against "real parts"
including the AD ones.
If you have enough gates in your FPGA, miraculous things can be
done. After all, "it's just software".
> and having done the BDR measurements (which
>I consider to be altogether basically unimportant in comparison to IP3,
>IP2, and IMD-DR), it appears they are good as seen in the ARRL lab
>results. Their results are consistent with my measurements.
>
>However, as I said, the proof is in the eating of the pudding. I have
>eval boards on the way.
>
>Bob
>
>
>
>
>
>Tayloe Dan-P26412 wrote:
>
> >>Let me say that for the future AD9954 and AD9958 just look spectacular.
> >>I have ordered the development boards for both of them. Especially on
> >>the low bands, either of these would give spur performance where the
> >>spurs are below the noise floor at all but sub multiples of the clock.
> >>It is my understanding that the 9958 has made spectacular strides in
> >>this "keeping the clock submultiples" off the output.
> >>
> >>
> >
> >The very first thing you need to do when evaluating a new part is to
> read the spec sheet. Once again, the spec sheet for the new AD9958 device
> can be found at:
> >
> >http://www.analog.com/UploadedFiles/Data_Sheets/383477232AD9958_prd.pdf
> >
> >Read it. See what it says. Spec sheets are normally optimistic, not
> pessimistic. The real world results are rarely much better than what is
> in the spec sheet.
> >
> >The close-in spur results specs are given on page 42. The specs for 1.1
> MHz are not significantly different than those for 15.1 MHz. The close
> in spurs at 1.1 MHz are 90 db down +/- 10 KHz, 88 db down +/- 50 KHz,
> very similar to the numbers that are shown for 15.1 MHz and the numbers
> at 40.1 MHz.
> >
> >These spurs would not be as much a problem if they were sparsely
> distributed. Again, the information is contained in the data sheet for
> us to get a handle on this. Page 12 shows the spur distribution for both
> 1.1 MHz (Figure 12) and 15.1 MHz (figure 15). In both cases, there are
> numerous spurs in the 80 to 90 db down region.
> >
> >What does this mean to the receiver?
> >
> >In the case of my NC2030, a very low power, high performance "hardware"
> SDR, the blocking dynamic range is 130 db at 10 KHz, and over 140 db at
> 20 KHz. On 20m, the sensitivity in a 500 Hz BW is -135 dbm. This means
> that a signal has to be over +5 dbm 20 KHz away to cause the receiver
> front end to go into compression, and -5 db when only 10 KHz away. If
> the LO has numerous spurs in the -80 to -90 db region and the sensitivity
> were -135 dbm, then signals 80 to 90 db higher than the receiver noise
> floor will get mixed on frequency. Thus, even though the receiver front
> end might be capable of rejecting signals up to +5 dbm 20 KHz away, a LO
> spur at 80 db down will cause a signal at only -55 dbm to appear on
> frequency as crud.
> >
> >This is not good. The LO limitations have now caused the receiver to
> give up 60 db of blocking dynamic range when this happens. It might not
> be so bad if we were only talking one close in spur, but from the sheets
> above, the spurs are numerous. In a contest weekend, we can have many
> signals mixed on frequency due to this effect, artificially raising the
> apparent noise floor of the band, and potentially masking the signals we
> want to hear.
> >
> >If you have a very clean signal generator such as an old HP8640B, you
> can go looking for these spurs yourself. Set the generator to a level
> 100 db above the noise floor of the signal and sweep it across a range 2
> to 100 KHz away from the receiver center frequency. You should actually
> see spurs pop up and move around as you sweep the generator across this
> region. Try a couple of different bands and a couple of different center
> frequencies and see what kind of variation you get.
> >
> >You need to understand what DDS chips were designed for in the first
> place to understand their performance limitations. These chips were
> designed originally for cellular telephone base stations. These base
> stations have a performance requirement that is quite different than what
> hams need. A cellular telephone base station has a group of mobiles that
> are in the range of its coverage. A closed loop power control mechanism
> is used between the base station and the mobile phone to reduce the
> phones power such that its signal is just sufficient for good
> communications with the base station. In CDMA phones (Verizon, Sprint,
> Alltel), the speech data is sent in 20msec data packets, and it is
> typical to set the mobile received power at the base station such that
> there is a 2% packet erasure rate.
> >
> >The bottom line is that to a cellular base station receiver, ***** all
> mobiles arrive at about the same power level *****. The base station
> does not have to typically worry about very strong signals adjacent to
> weak signals. In such an environment, spurs only 80 db down is perfectly
> fine. Likewise on transmit, the base stations need spectral purity of (I
> think) 70 db down. This is much more stringent than ham transmitters,
> but again, spurs 80 db down or more are ok.
> >
> >Thus DDSs work great for the intended application, cellular base
> stations. However, ham receivers do not face signals that are all
> uniform in strength. We have very weak signals right next to very strong
> ones. Thus ideally we would like to have high sensitivity in order to
> hear the weak ones combined with high adjacent signal rejection, thus
> blocking and IP3 specs.
> >
> >In my case, IP3 is more important than blocking specs. A receiver will
> distort 20+ db before it overloads. However, I do not have any big gun
> stations in my area and this is not the case for other folks. For anyone
> that has a big gun station near by, blocking performance is also very
> important. All it takes is one very large signal to cause a receiver a
> problem.
> >
> >If all you want is 80 to 90 db of dynamic range under crowded band
> conditions, a DDS is fine. If you look at QST receiver reviews, there
> are a lot of good receivers that have this kind of IP3 performance,
> although it would be kind of poor blocking performance. This level of
> performance would definitely limit a great rig such as a K2, an Orion, or
> in my case the NC2030. If you want to see more information on the
> NC2030, go to www.norcalqrp.org/nc2030.htm and read some of the
> presentations there.
> >
> >Any body can throw stones at a receiver design. It is a very complex
> subject with lots of tradeoffs. Thus, let me offer a few suggestions on
> how to improve the LO chain performance. First, I would suggest going to
> a PLL VCO design that tunes in something like 10 KHz steps. With the SDR
> approach, sub Hz tuning is simply not necessary since the SDR software
> can tune to all the frequencies in between. Since the PLL reference will
> be a clean crystal oscillator, the VCO should inherent some of the
> reference oscillators clean phase noise characteristics. It will have
> spurs at the reference frequency harmonics, but these will be much fewer
> than in DDS and can be greatly minimized by known, good PPL design. The
> PLL approach is common in high performance ham rigs.
> >
> >Another possibility that came to mind is coming up with a set of "magic"
> DDS frequencies. If the rig really only needs to have 10 KHz LO tuning
> steps, you could take advantage of the fact that not all DDS tuning steps
> have the same spur problems. It seems to me that there is a subset of
> DDS tuning setting that may have very few (if any) close in spurs. If
> some how these tuning setting could be identified, you could get the best
> of both worlds, great phase noise, and minimal close in spurs. It may be
> that optimum settings could be determined mathematically. I am not sure
> this approach is really possible, but it seems worth investigating.
> >
> >It is also very neat that something as simple as a "soft rock" has the
> potential of zero problems with spurs or phase noise because of the use
> of a crystal oscillator.
> >
> >Fun stuff!
> >
> >- Dan, N7VE
> >
> >_______________________________________________
> >FlexRadio mailing list
> >FlexRadio@flex-radio.biz
> >http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz
> >
> >
> >
>
>
>_______________________________________________
>FlexRadio mailing list
>FlexRadio@flex-radio.biz
>http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz
James Lux, P.E.
Spacecraft Radio Frequency Subsystems Group
Flight Communications Systems Section
Jet Propulsion Laboratory, Mail Stop 161-213
4800 Oak Grove Drive
Pasadena CA 91109
tel: (818)354-2075
fax: (818)393-6875
James Lux, P.E.
Spacecraft Radio Frequency Subsystems Group
Flight Communications Systems Section
Jet Propulsion Laboratory, Mail Stop 161-213
4800 Oak Grove Drive
Pasadena CA 91109
tel: (818)354-2075
fax: (818)393-6875
