Announcement

Collapse
No announcement yet.

The Tayloe Mixer: A Low Noise Solution

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #76
    Ok Im back - Can one of you recommend a decent mixed signal sim ?


    So at least one of us can duplicate Carls efforts.


    Clearly there is a small advantage here in one of the permutations - Carl is playing his cards close to his chest..


    If I had a random stab in the dark I would go for ~Carls full wave clocked at 4x or somesuch..

    I want to run something tonight but ltspice baulked at anything more than one Fet switcher integrator.


    Carl is your sim an internal app? or is it open to all.

    Steve

    Comment


    • #77
      Originally posted by golfnut View Post
      Ok Im back - Can one of you recommend a decent mixed signal sim ?


      So at least one of us can duplicate Carls efforts.


      Clearly there is a small advantage here in one of the permutations - Carl is playing his cards close to his chest..


      If I had a random stab in the dark I would go for ~Carls full wave clocked at 4x or somesuch..

      I want to run something tonight but ltspice baulked at anything more than one Fet switcher integrator.


      Carl is your sim an internal app? or is it open to all.

      Steve
      Am I wrong... it seems to me the Tayloe and "full-wave" demods can be configured mathematically the same with enough components -- Tayloe just integrating quarter-wave at a time, and the full-wave integrating half-wave at a time. You can make the sums equal if you include the same pieces, I'd think.

      -SB

      Comment


      • #78
        Originally posted by Carl-NC View Post
        There are no overlapping ON periods in the other two, either.
        Interesting. I thought at least half wave must have had overlapping ON periods. It is not that straightforward on your schematics.
        Anyway, those resistors are not required.

        Comment


        • #79
          Tayloe and "full-wave" demods can be configured mathematically the same with enough components


          Sorry Im not clear on this can u explain it a different way pls

          Just had another go at basic LT spice - it cant handle it. Anyone got a sim thats powerlul enough and wants to save the world with a few models like Carls - so we can compare results with oregon..


          S

          Comment


          • #80
            Ah, so you ran into a convergence problem. It happens.

            In most of the cases you just have to switch to the "Alternative" solver and it runs fine. It produces somewhat hairier diagrams, but just runs through where standard solver gets stuck.
            In LTspice go to Tools->Control Panel, you'll find a Spice tab, and in there you'll find Solver. There you pick alternative, and in most cases it works. Normal is supposed to be more accurate, but heck, models are not that perfect either.

            There are some other tricks, but I never needed to revert to them.

            Comment


            • #81
              Hi all,

              removing (reducing) the input resistance to the mixer (integrator) should be of more benefit: -> low thermal noise.

              But I think, the resistor noise isn't much critical in MD applications.

              Aziz

              BTW, I prefer digital (software) demodulation of the signals.

              Comment


              • #82
                Originally posted by golfnut View Post
                Carl is your sim an internal app? or is it open to all.
                I'm using the now-defunct Easy Spice, which I got with Easy-PC. It's a nice Spice implementation, not sure why they dropped it. Unfortunately, it can't do transient noise analysis (truly necessary for mixer sims), but neither can any other Spice. I would need Spectre RF for that, and I ain't got that kind of money.

                Originally posted by simonbaker View Post
                Am I wrong... it seems to me the Tayloe and "full-wave" demods can be configured mathematically the same with enough components -- Tayloe just integrating quarter-wave at a time, and the full-wave integrating half-wave at a time. You can make the sums equal if you include the same pieces, I'd think.
                Yup.

                Originally posted by Davor View Post
                Interesting. I thought at least half wave must have had overlapping ON periods. It is not that straightforward on your schematics.
                Anyway, those resistors are not required.
                True, but they serve the Useful Purpose of comparing drive strength. I adjusted the R's to equalize drive strength, so based on the values I ended up with you can guess that the full-wave method had the strongest drive, with half-wave and Tayloe having equivalent lower drive (Tayloe has a lower R, but has twice the outputs).

                Originally posted by Davor View Post
                In most of the cases you just have to switch to the "Alternative" solver and it runs fine. It produces somewhat hairier diagrams, but just runs through where standard solver gets stuck.
                In Berkeley Spice, trapezoidal is the default integration method, which gives smoother results but can hit convergence problems. Gear integration is the alternative, switch to it using .options method=gear. Some Spice have Gear2 or Gear3 options. Also loosen up the tols to improve convergence, especially chgtol in transient sims.

                Originally posted by Aziz View Post
                But I think, the resistor noise isn't much critical in MD applications.
                I tend to agree.

                BTW, I prefer digital (software) demodulation of the signals.
                Except the necessary processing horsepower goes up exponentially!

                - Carl

                Comment


                • #83
                  Originally posted by Carl-NC View Post
                  I would need Spectre RF for that, and I ain't got that kind of money.
                  Apparently HSPICE also has Transient Noise Analysis.
                  But you still need a lot of dosh for that as well.

                  Comment


                  • #84
                    This is a cct I was trying anyone care to run it - I was hoping for 1Hz out.

                    Prob is takes all day and max's pc and sim - pants
                    Attached Files

                    Comment


                    • #85
                      Same here, but the moment you assign those NJFs some real component designation (say 2n3819) it finishes in a jiffy. There are several other problems using the values specified, but you'll come up wit them eventually.

                      Comment


                      • #86
                        It looks like this thread has faded now.

                        Was there any real conclusion of X over Y

                        S

                        Comment


                        • #87
                          Sorry about some thread necromancy, but I believe there are a few things to point out which weren't listed in the thread.

                          The "Tayloe" modulator/demodulator, or quadrature switching mod/demod - which has been around a bit longer than the radio amateur's paper which attributed his name to it in those circles - is configured for the purpose of dealing with quadrature signals. The reason for using four separate 90 deg phases per switch instead of two overlapping 90 deg phases is ease of coupling with a phase shifting network necessary for image frequency cancellation, and isolating the relatively low impedance nodes from each other, as the driving impedance is comparatively high. The divider circuits used will (with some attention) deliver far more precise 90 degree shifts over a wider frequency range than passive networks.

                          Why is it used right after the antenna network? It is usually used ahead of amplifiers due to high dynamic range of mixing, which is often required for radio reception in presence of powerful signals. Other active components may sport relatively small dynamic ranges, where stronger signals may swamp out weaker ones by compression. The touted low 1-2dB input loss is due to using a high performance bus switch component. "Ordinary" 4000/74HC analog switches will have noticeably higher losses, so there's no inherent magic to the quadrature switching design.

                          This is as far as I know from reading up on the topic. Also for MD uses, we might not always want a precise 90 deg shift between the demodulated phases. With a low impedance drive (input amplifier's output) using two 180 degree switches should not be an issue, although peak level is not as high - but after the input amplifier this should not pose an issue either.

                          Comment


                          • #88
                            Actually there is more to it. Recently I learned that there may be much more to it, and some of these ideas I gave at LF project topic.

                            First use of a QSD by means of bilateral switches implementation in form of a MUX that is now mostly attributed to Tayloe was published in September 1990 issue of QEX in D. H. van Graas PA0DEN article "The Fourth Method: Generating and Detecting SSB Signals", at least the first that I know of.

                            As for the conversion loss, it is related to integration of the waveform during sample period, normalised for the period. The switching mixers with duty cycle of 50% for unbalanced mixers, and twice 50% for balanced mixers, is calculated as an integral of a sinewave from 0 to π, or as a sum of integrals in periods 0 to π and π to 2π in balanced case. For the balanced case it results 2/π or -3.9dB.
                            For QSD conversion loss is better as the integration happens for the top of a sinewave between 45°and 135° and again at 225° and 315°. Theoretical conversion loss in this case is only -0.9dB for a balanced situation. As the noise matching at low frequencies does not require impedance matching, there are virtually no conduction losses in switches so 1-2dB loss is achievable.

                            As for fine tuning of the phases you can always revert to addition/subtraction of I and Q channels (as pioneered by Fisher) to achieve any angle imaginable, and even more interesting - completely independent of a Tx frequency.

                            I intend to pursue this particular route in my LF project (to be). I realised that there are some very tempting approaches involving QSD as a chopper front-end that will lead to DC accurate gain blocks and freedom from 1/f noise.

                            Comment


                            • #89
                              True, the peak level is not as high. Also, Fisher is prolific among phase tweaking and summing designs and patents.

                              Comment

                              Working...
                              X