Loading ...
Sorry, an error occurred while loading the content.
 

Re: [spectrumanalyzer] Re: Filter issues/pll choice

Expand Messages
  • william sprowls
    Hello Swetterlin, I ll make a quick comment on your observations. The DDS filter should be a narrow as possible, while allowing the frequency movement
    Message 1 of 11 , Jun 11, 2005
      Hello Swetterlin,
      I'll make a quick comment on your observations. The
      DDS filter should be a narrow as possible, while
      allowing the frequency movement necessary to drive the
      PLL 1. The AD9850 is only a 10 bit D/A process and
      the spurs are numerous and fairly high in amplitude
      (-60 dBc). The newer and fancier Analog Devices DDS's
      have 12 bits or greater D/A. Their spurs are the same
      in frequency, but much lower in amplitude. You can
      use a 10.7 MHz filter with a bandwidth of 15 KHz as a
      DDS output filter. I have tried one in my SA's and it
      does allow more spur noise to effect PLL 1. With this
      wider DDS filter, you can use a higher phase detector
      frequency for PLL 1, about 920 KHz. This will give
      you better phase noise, but, at the expense of larger
      spurs.
      You are absolutely correct on the use of a very
      narrow filter to determine the resolution bandwidth of
      a spectrum analyzer. It will create "ringing" if
      swept to fast. In the case of the Modularized
      Spectrum Analyzer, the sweep speed can be adjusted to
      a slow rate to prevent ringing. It is of more concern
      that one might miss signals when sweeping a wide
      range. Remember, the MSA does not really sweep, but
      rather, steps in frequency increments. If the
      increment is larger than the bandwidth of the final
      filter, some frequencies might be "skipped over". It
      would be preferable to use a wide bw final filter when
      sweeping large segments, and a narrow bw filter when
      the sweep is very narrow. There is a software
      provision in the MSA to control up to 4 sets of
      switches and would change those final filters
      automatically.
      Cheers, Scotty

      --- swetterlin <swetterlin@...> wrote:
      > It seems to me that people are overly fixated on
      > narrow bandwidths.
      > If you are talking about filtering the DDS output
      > around 10.7MHz, it
      > appears that the major concern is close-in spurs,
      > but they don't seem
      > to be all that close in--I don't see why a 15MHz
      > bandwidth ceramic
      > filter would not do fine. If you are talking about
      > the final filter,
      > which determines the resolution of the final output,
      > a narrow filter
      > does have its place, but it also creates problems
      > with sweep speed.
      > If you sweep through the frequencies with a very
      > narrow filter, you
      > get a ringing effect that turns your narrow spike
      > into a little
      > balloon with side balloons, so the narrow filter is
      > only useful for
      > display of narrow or very slow sweeps. It seems to
      > me that the ideal
      > solution would be to use a pair of rf switches to
      > switch between a
      > wide and narrow filter--Analog Devices now has some
      > great IC rf
      > switches which are cheap and one-input-TTL
      > switchable. One of their
      > configurations is SPDT; one of these at the filter
      > input and one at
      > the output could switch the filters with over 60db
      > isolation, and
      > could be computer controlled with one bit.




      __________________________________
      Discover Yahoo!
      Find restaurants, movies, travel and more fun for the weekend. Check it out!
      http://discover.yahoo.com/weekend.html
    • swetterlin
      Hi Scotty, Anything that shows up as a spur in the final output would be an irritation, even if small. But when the DDS output is squared off, any amplitude
      Message 2 of 11 , Jun 12, 2005
        Hi Scotty,

        Anything that shows up as a spur in the final output would be an
        irritation, even if small. But when the DDS output is squared off,
        any amplitude effect of the spurs is gone and we are left only with
        jitter in the square wave, which is analogous to phase noise. Is
        there any simple way to approximate whether significant phase noise
        results from a spur of a given level?
        On the final filter,
        You make a good point about the possibility of simply skipping over
        an important frequency. I note that commercial SAs offer resolution
        bandwidths as high as 1MHz for use in broad sweeps. We could use an
        ADG904 1-of-4 multiplexer ($3.70 at Digikey) to direct the output to
        one of four filters, and another to direct one of the filter outputs
        to further processing. But each filter would have its own insertion
        loss so maybe some calibration is necessary (particularly when
        adapting the SSA for use as a vector analyzer--though in that case our
        tuning is always directly "on frequency").

        --- In spectrumanalyzer@yahoogroups.com, william sprowls
        <wsprowls@y...> wrote:
        > Hello Swetterlin,
        > I'll make a quick comment on your observations. The
        > DDS filter should be a narrow as possible, while
        > allowing the frequency movement necessary to drive the
        > PLL 1. The AD9850 is only a 10 bit D/A process and
        > the spurs are numerous and fairly high in amplitude
        > (-60 dBc). The newer and fancier Analog Devices DDS's
        > have 12 bits or greater D/A. Their spurs are the same
        > in frequency, but much lower in amplitude. You can
        > use a 10.7 MHz filter with a bandwidth of 15 KHz as a
        > DDS output filter. I have tried one in my SA's and it
        > does allow more spur noise to effect PLL 1. With this
        > wider DDS filter, you can use a higher phase detector
        > frequency for PLL 1, about 920 KHz. This will give
        > you better phase noise, but, at the expense of larger
        > spurs.
        > You are absolutely correct on the use of a very
        > narrow filter to determine the resolution bandwidth of
        > a spectrum analyzer. It will create "ringing" if
        > swept to fast. In the case of the Modularized
        > Spectrum Analyzer, the sweep speed can be adjusted to
        > a slow rate to prevent ringing. It is of more concern
        > that one might miss signals when sweeping a wide
        > range. Remember, the MSA does not really sweep, but
        > rather, steps in frequency increments. If the
        > increment is larger than the bandwidth of the final
        > filter, some frequencies might be "skipped over". It
        > would be preferable to use a wide bw final filter when
        > sweeping large segments, and a narrow bw filter when
        > the sweep is very narrow. There is a software
        > provision in the MSA to control up to 4 sets of
        > switches and would change those final filters
        > automatically.
        > Cheers, Scotty
        >
        > --- swetterlin <swetterlin@m...> wrote:
        > > It seems to me that people are overly fixated on
        > > narrow bandwidths.
        > > If you are talking about filtering the DDS output
        > > around 10.7MHz, it
        > > appears that the major concern is close-in spurs,
        > > but they don't seem
        > > to be all that close in--I don't see why a 15MHz
        > > bandwidth ceramic
        > > filter would not do fine. If you are talking about
        > > the final filter,
        > > which determines the resolution of the final output,
        > > a narrow filter
        > > does have its place, but it also creates problems
        > > with sweep speed.
        > > If you sweep through the frequencies with a very
        > > narrow filter, you
        > > get a ringing effect that turns your narrow spike
        > > into a little
        > > balloon with side balloons, so the narrow filter is
        > > only useful for
        > > display of narrow or very slow sweeps. It seems to
        > > me that the ideal
        > > solution would be to use a pair of rf switches to
        > > switch between a
        > > wide and narrow filter--Analog Devices now has some
        > > great IC rf
        > > switches which are cheap and one-input-TTL
        > > switchable. One of their
        > > configurations is SPDT; one of these at the filter
        > > input and one at
        > > the output could switch the filters with over 60db
        > > isolation, and
        > > could be computer controlled with one bit.
        >
        >
        >
        >
        > __________________________________
        > Discover Yahoo!
        > Find restaurants, movies, travel and more fun for the weekend. Check
        it out!
        > http://discover.yahoo.com/weekend.html
      • John Miles
        That small irritation is the end of the world in the spectrum-analyzer business. Phase noise has absolutely nothing to do with spurs, and squaring the signal
        Message 3 of 11 , Jun 12, 2005
          That 'small irritation' is the end of the world in the spectrum-analyzer business.
           
          Phase noise has absolutely nothing to do with spurs, and squaring the signal cannot separate the spurs' "amplitude effect" from their frequency-domain representation.
           
          In fact, squaring the output of the DDS does nothing to reduce either noise *or* spurious content, except to the extent that the PLL chip is happier with a strong logic-level signal rather than an analog signal that it has to amplify and trigger on.  (Basically, a dedicated external comparator is likely to be better than the ones that are bundled for 'free' with PLL and DDS chips.)
           
          What will *really* bake your noodle is thinking about why squaring the output of the DAC in the DDS chip is not the same as just feeding the high bit of the phase accumulator directly to the PLL chip!
           
          -- john KE5FX
          -----Original Message-----
          From: spectrumanalyzer@yahoogroups.com [mailto:spectrumanalyzer@yahoogroups.com]On Behalf Of swetterlin
          Sent: Sunday, June 12, 2005 9:49 AM
          To: spectrumanalyzer@yahoogroups.com
          Subject: [spectrumanalyzer] Re: Filter issues/pll choice

          Hi Scotty,

          Anything that shows up as a spur in the final output would be an
          irritation, even if small.  But when the DDS output is squared off,
          any amplitude effect of the spurs is gone and we are left only with
          jitter in the square wave, which is analogous to phase noise.  Is
          there any simple way to approximate whether significant phase noise
          results from a spur of a given level?
          On the final filter,
          You make a good point about the possibility of simply skipping over
          an important frequency.  I note that commercial SAs offer resolution
          bandwidths as high as 1MHz for use in broad sweeps.  We could use an
          ADG904 1-of-4 multiplexer ($3.70 at Digikey) to direct the output to
          one of four filters, and another to direct one of the filter outputs
          to further processing.  But each filter would have its own insertion
          loss so maybe some calibration is necessary (particularly when
          adapting the SSA for use as a vector analyzer--though in that case our
          tuning is always directly "on frequency").
        • swetterlin
          In trying to reply, I sent a blank message--sorry about that. I didn t mean to say the irritation was small, rather that a small spike on the output display is
          Message 4 of 11 , Jun 12, 2005
            In trying to reply, I sent a blank message--sorry about that.
            I didn't mean to say the irritation was small, rather that a small spike on the output display
            is an irritation.

            But I disagree on the separation of amplitude and frequency effects of spurs. These spurs
            represent an increase in noise at particular frequencies, though not always random
            noise--e.g. they can represent amplitude modulation. When any signal with noise is
            squared off by a comparator, every cycle is identical to the next except possibly in the
            positioning of its vertical transitions, which constitutes jitter. For example, a sine wave
            with heavy noise at its peaks, but no noise near its zero crossing, when squared off will be
            perfectly noise free (except for any new noise introduced by the comparator). On the
            other hand, even a small amount of noise near the zero crossing will affect the timing of
            the comparator's switching, and create lots of jitter. The primary function of the
            comparator is to eliminate all noise other than that which affects the comparator's
            transition timing.

            --- In spectrumanalyzer@yahoogroups.com, "John Miles" <jmiles@p...> wrote:
            > That 'small irritation' is the end of the world in the spectrum-analyzer
            > business.
            >
            > Phase noise has absolutely nothing to do with spurs, and squaring the signal
            > cannot separate the spurs' "amplitude effect" from their frequency-domain
            > representation.
            >
            > In fact, squaring the output of the DDS does nothing to reduce either noise
            > *or* spurious content, except to the extent that the PLL chip is happier
            > with a strong logic-level signal rather than an analog signal that it has to
            > amplify and trigger on. (Basically, a dedicated external comparator is
            > likely to be better than the ones that are bundled for 'free' with PLL and
            > DDS chips.)
            >
            > What will *really* bake your noodle is thinking about why squaring the
            > output of the DAC in the DDS chip is not the same as just feeding the high
            > bit of the phase accumulator directly to the PLL chip!
            >
            > -- john KE5FX
            > -----Original Message-----
            > From: spectrumanalyzer@yahoogroups.com
            > [mailto:spectrumanalyzer@yahoogroups.com]On Behalf Of swetterlin
            > Sent: Sunday, June 12, 2005 9:49 AM
            > To: spectrumanalyzer@yahoogroups.com
            > Subject: [spectrumanalyzer] Re: Filter issues/pll choice
            >
            >
            > Hi Scotty,
            >
            > Anything that shows up as a spur in the final output would be an
            > irritation, even if small. But when the DDS output is squared off,
            > any amplitude effect of the spurs is gone and we are left only with
            > jitter in the square wave, which is analogous to phase noise. Is
            > there any simple way to approximate whether significant phase noise
            > results from a spur of a given level?
            > On the final filter,
            > You make a good point about the possibility of simply skipping over
            > an important frequency. I note that commercial SAs offer resolution
            > bandwidths as high as 1MHz for use in broad sweeps. We could use an
            > ADG904 1-of-4 multiplexer ($3.70 at Digikey) to direct the output to
            > one of four filters, and another to direct one of the filter outputs
            > to further processing. But each filter would have its own insertion
            > loss so maybe some calibration is necessary (particularly when
            > adapting the SSA for use as a vector analyzer--though in that case our
            > tuning is always directly "on frequency").
          • Scotty
            Hi All, In the case of changing final filters to change the resolution bandwidth of the spectrum analyzer: It is a given that each different filter will have
            Message 5 of 11 , Jun 12, 2005
              Hi All,
              In the case of changing final filters to change the resolution
              bandwidth of the spectrum analyzer: It is a given that each
              different filter will have different insertion loss. In expensive
              spectrum analyzers, each filter usually has a dedicated, "make up"
              amplifier (or attenuation pad) to maintain the same loss (or gain) in
              each filter bank. I have written the software in such a way as to
              have a separate calibration for each filter, so that the vertical
              scale will be correct for each filter. Very cheap way to do it but
              you will notice that the scale values will change with each different
              filter. The "software" page on the web site tells you how to
              calibrate for each filter.
              Spurs and noise: Well yes, there is a big difference between the
              two. In the DDS, the digital to analog accumulator is the cause of
              spurs and aliases. The action of the spurs are a direct correlation
              between the clock frequency and the output frequency. The spurs can
              be thought of and can be calculated as mixing products of the two.
              For example, with a clock of 100 MHz and an output of 20 MHz there
              will be signals created at clock-output(80 MHz) and clock+output(120
              MHz). These are actually aliases, but can be thought of as spurs.
              There will also be signals created at N times clock +/- M times
              output, where N and M are integer numbers between 1 and infinity.
              These signals are usually referred to as spur frequencies. Any low
              pass filter will get rid of alias frequencies and spurs that are
              greater than the cut-off of the filter. However, let us look at the
              spur that is generated at the frequency: clock - 4 times output (100-
              80). That is equal to 20 MHz, the exact same frequency as the output
              frequency. That doesnt sound like a big deal until you change the
              wanted output frequency to 20.000100 MHz (100 Hz higher). The spur
              that is created is: clock - 4 times output (100-80.000400) =
              19.999600 MHz. This spur is 500 Hz below the output frequency.
              Using even a narrow 2 KHz bandwidth filter in the DDS output will not
              get rid of this nasty little spur. This is why I give a warning: do
              not to use a DDS output frequency that will be close to a sub
              multiple of the clock frequency (1/5 of clock, etc). Even with that
              closely followed, there will be unwanted spurs within the bandwidth
              of your DDS output filter, due to higher order intermodulation
              products. The wider the filter, the more spurs will be there.
              Granted, they may be very low in amplitude but they are still a pain
              in the backside. That is why to use a narrow as filter as possible.
              In many cases, such as radio receivers, interferring spurs that are
              60 dB down may not be heard in the audio demodulation. But as John
              says, it is a visual nightmare in a spectrum analyzer.
              No amount of squaring the signal of interest will eliminate close-in
              spurs. They will, however, be somewhat attenuated due to the
              compressing effect of a limiting amplifier (squarer).
              Oddly enough, generated spurs have their own jitter (and
              consequently, phase noise). However, spurs that are at least 30 dB
              down will have no additive effect on phase noise of the signal of
              interest. Analog Devices says the AD9850 spurs will be at least 60
              dB down. Most of the spurs I have measured have been about 75 dB
              down.
              Cheers, Scotty
            • swetterlin
              Hi Scotty Thanks for the info. That clarifies the spur thing a great deal. By the way, if anyone is interested in experimenting with a DDS, minikits.com sells
              Message 6 of 11 , Jun 13, 2005
                Hi Scotty
                Thanks for the info. That clarifies the spur thing a great deal. By
                the way, if anyone is interested in experimenting with a DDS,
                minikits.com sells a DDS board with on board PIC micro control so you
                can play with it without a PC attached. I assembled mine but have not
                yet experimented with it. Not cheap (about $100 plus LCD display and
                keypad) but I expect it to be very educational.

                --- In spectrumanalyzer@yahoogroups.com, "Scotty" <wsprowls@y...> wrote:
                > Hi All,
                > In the case of changing final filters to change the resolution
                > bandwidth of the spectrum analyzer: It is a given that each
                > different filter will have different insertion loss. In expensive
                > spectrum analyzers, each filter usually has a dedicated, "make up"
                > amplifier (or attenuation pad) to maintain the same loss (or gain) in
                > each filter bank. I have written the software in such a way as to
                > have a separate calibration for each filter, so that the vertical
                > scale will be correct for each filter. Very cheap way to do it but
                > you will notice that the scale values will change with each different
                > filter. The "software" page on the web site tells you how to
                > calibrate for each filter.
                > Spurs and noise: Well yes, there is a big difference between the
                > two. In the DDS, the digital to analog accumulator is the cause of
                > spurs and aliases. The action of the spurs are a direct correlation
                > between the clock frequency and the output frequency. The spurs can
                > be thought of and can be calculated as mixing products of the two.
                > For example, with a clock of 100 MHz and an output of 20 MHz there
                > will be signals created at clock-output(80 MHz) and clock+output(120
                > MHz). These are actually aliases, but can be thought of as spurs.
                > There will also be signals created at N times clock +/- M times
                > output, where N and M are integer numbers between 1 and infinity.
                > These signals are usually referred to as spur frequencies. Any low
                > pass filter will get rid of alias frequencies and spurs that are
                > greater than the cut-off of the filter. However, let us look at the
                > spur that is generated at the frequency: clock - 4 times output (100-
                > 80). That is equal to 20 MHz, the exact same frequency as the output
                > frequency. That doesnt sound like a big deal until you change the
                > wanted output frequency to 20.000100 MHz (100 Hz higher). The spur
                > that is created is: clock - 4 times output (100-80.000400) =
                > 19.999600 MHz. This spur is 500 Hz below the output frequency.
                > Using even a narrow 2 KHz bandwidth filter in the DDS output will not
                > get rid of this nasty little spur. This is why I give a warning: do
                > not to use a DDS output frequency that will be close to a sub
                > multiple of the clock frequency (1/5 of clock, etc). Even with that
                > closely followed, there will be unwanted spurs within the bandwidth
                > of your DDS output filter, due to higher order intermodulation
                > products. The wider the filter, the more spurs will be there.
                > Granted, they may be very low in amplitude but they are still a pain
                > in the backside. That is why to use a narrow as filter as possible.
                > In many cases, such as radio receivers, interferring spurs that are
                > 60 dB down may not be heard in the audio demodulation. But as John
                > says, it is a visual nightmare in a spectrum analyzer.
                > No amount of squaring the signal of interest will eliminate close-in
                > spurs. They will, however, be somewhat attenuated due to the
                > compressing effect of a limiting amplifier (squarer).
                > Oddly enough, generated spurs have their own jitter (and
                > consequently, phase noise). However, spurs that are at least 30 dB
                > down will have no additive effect on phase noise of the signal of
                > interest. Analog Devices says the AD9850 spurs will be at least 60
                > dB down. Most of the spurs I have measured have been about 75 dB
                > down.
                > Cheers, Scotty
              Your message has been successfully submitted and would be delivered to recipients shortly.