Connect with us

analogue PLL

Discussion in 'Electronic Design' started by colin, Sep 1, 2007.

Scroll to continue with content
  1. colin

    colin Guest

    Hi,
    Im looking at the idea of using an analogue PLL
    to add a fixed offset to a variable frequency.

    but using a ramped offset to the phase error in the feedback loop
    instead of using a mixer to add the frequency offset wich has problems.

    to avoid problems at the limits of the phase error range
    it would probably be best to do this in increments of 90'
    by some simple divide by 4 logic.

    the input frequency is ramped from 1-2ghz,
    and the offset is ~500khz.

    alternatives such as fractional divison pll and dds arnt without there cons.

    is this something that has been done before ?

    Colin =^.^=
     
  2. MooseFET

    MooseFET Guest

    I have done this at lower frequencies. It ends up being quite a bit
    like the method with mixers. That gain you want to apply to each
    phase detector's output is a sine function. This takes the gain
    smoothly through zero as the phase detector goes through its maximum
    value.
     
  3. Phil Hobbs

    Phil Hobbs Guest

    Offset PLLs are a piece of cake. The reason is that you can filter off
    all the mixer spurs in the control voltage, leaving the output very
    clean. The biggest problem is that when using sinusoidal PDs, offset
    PLLs can lock up on the wrong sideband. PFDs solve this problem (we had
    a thread on this some time back), but aren't that easy to do at 2 GHz.

    The usual approach is to use an auxiliary mixer driven in quadrature, so
    you can distinguish +pi/2 from -pi/2 (and hence USB from LSB) by looking
    at the sign of the output from the quadrature mixer.

    Cheers,

    Phil Hobbs
     
  4. Chris Jones

    Chris Jones Guest

    As the other posters suggested, yes it has been done before, the offset loop
    is one of the standard methods used in GSM mobile phone transmitters.

    Basically what you are doing is putting a mixer in the feedback path to the
    divider (or after the divider if you prefer, but that's more noisy).

    If you tell us what you are trying to achieve, and your performance
    requirements, then we may be able to suggest the best approach.

    Chris
     
  5. colin

    colin Guest

    offset PLL you refer to is that as a specific/distinct topology/layout ?

    I dont know if you refering to some other threads where ive explored posible
    ways to do this,
    the thing that makes it difficult is the closeness of the frequencies they
    tend to lock together.

    so the loop needs to respond quite quickly.

    I was thinking of using a ecl prescaler to bring it down to the 100mhz
    region,
    and a digital phase detector so the phase signal is idealy linear to phase.

    Further thinking I realise I can use an xor type PD, and a triangle
    waveform,
    ofc I would need to reverse the polarity of the error amplifier,
    this would avoid discontinuties of a ramp but im not clear in my mind how it
    would behave for locking etc.

    quadrature is difficult to generate well over a wide range at that
    frequency,
    with a good quadrature mixer system I could generate the 2nd signal directly
    however ive tried this and the result isnt a very clean signal,
    it probably would be ok at lower frequencies or narrow range or if I paid a
    lot of attention to layout,
    although this could be cleaned up with a PLL.

    you can get ecl PFD, the problem is here that they will sense one frequency
    is always higher.
    it would be posible to use combination of FF and xor type to have 1 PD for
    each quadrant.

    a gate array might be usefull here, it could generate the appropriate pwm
    for offset frequency waveform too.

    Colin =^.^=
     
  6. colin

    colin Guest

    laser modulation frequency is ramped from ~1ghz to ~2ghz quite quickly,
    reflected signal is mixed with suitable LO to acheive ~500khz IF
    not much higher as its an optical mixer.

    phase versus frequency gives good corelation to reflected distance.
    => heterodyne lidar.

    some error in frequency can be compensated for,
    but theres a fair bit of feedthrough of any sidbands in the LO or RF signal.

    Colin =^.^=
     
  7. Andrew Holme

    Andrew Holme Guest

    The earliest fractional-N synthesizers (those which used analogue
    compensation) worked by applying a ramped offset to the phase detector
    output e.g. DANA Digiphase.
     
  8. Tam/WB2TT

    Tam/WB2TT Guest

    If he is doing the sweep by changing the value of some /N counter, he could
    use a second PLL whose input is the original binary frequency value plus
    some offset obtained from a digital full adder. You add a constant
    proportional to the IF frequency.

    Also, he might be able to use a 0 frequency IF.

    Tam
     
  9. Chris Jones

    Chris Jones Guest

    Could you give a little more detail? (I am unfamiliar with the system.)

    As I understand it, you are generating a modulating signal swept from 1GHz
    to 2GHz as well as an LO signal that is always 500kHz above (or below) the
    modulating signal. You then get some signal from a photodiode and mix this
    with the LO signal to get something around 500kHz, but slightly higher or
    lower because the received modulation is slightly higher or lower than the
    transmitted modulation signal. Is that correct?

    You mentioned in another post that there might be problems with the two
    signals (LO and modulating signal) locking together. In my experience this
    would not be a problem as long as you have enough money and space to place
    the oscillators in separate shielding enclosures, with individual power
    supply regulation and filtering, and individual isolation buffers to
    prevent RF from one oscillator getting back into the enclosure of the other
    oscillator. You will find that having a wide loop bandwidth for the loop
    that locks the two oscillators 500kHz apart will be very beneficial. You
    want it to have plenty of gain left at 500kHz.

    It might conceivably be easier to build something where the difference
    between the modulating and LO signal is not a constant offset, but rather
    where the ratio between the modulating and LO signals is constant. e.g.
    modulating signal sweeps 1GHz to 2GHz, LO sweeps 1.0005 to 2.001GHz. This
    way, you could just use two wide bandwidth frac-N synths which you program
    and lock with (say) a 10MHz reference frequency from a DDS, and then sweep
    the reference up to 20MHz, keeping the two synths locked the whole time.
    You would have to deal with the variable IF frequency somehow, perhaps in
    DSP.

    Chris
     
  10. colin

    colin Guest

    yes thats pretty much spot on, the reflected laser signal is mixed with
    the LO in a APD with the LO modulating the HV bias,
    this cuases multiplication inside the gap of the APD wich works very well as
    a mixer,
    the 500khz signal is very strong compared to say a signal converted at
    100mhz,
    as the RF signal mixes with the LO before it is attenuated by the junction
    capacitance,
    yet still has the benefits of large phase change with small distance change.
    as the RF frequency is swept the phase of the reflection changes
    proportional to distance
    the total phase change over the sweep represents the distance.

    the output from the mixer changes by the same phase,
    relative to the offset between RF and LO.

    500khz is the minimum desirable offset as
    the output lets through all sorts of crap from flourescent lights etc,

    so to have a DC IF as Tam suggested is interesting but would need
    something like a double balanced or quadrature mixer to properly reject
    this,
    im not sure how you would arange this, maybe a custom device?
    ot two APD with out of phase LO ?
    the APD are quite expensive it would be cool to think of a way to get a
    cheap
    photodiode to mix optically.

    to convert the RF signal directly to electrical
    results in far to poor SNR.

    although it is posible to tune the detector output
    this makes it dificult for wideband
    however it would allow me to use a 10mhz IF wich
    I have considered wich would make the signal generation quite easy.
    yes maybe but I was looking for s simpler way than spending lots of money or
    time.
    to have gain at 500khz means having quite a high comparison frequency,
    this makes it dificult to find a N/R ratio wich gives 500khz offset.

    the oscilators were moved to seperate boards with triple isolation buffers,
    the PLL,VCO,laser and driver were made into a very tiny well sheilded
    module.
    using PHEMT for vcos driven very hard made it considerably better still,
    but the mixed signal was still very distorted,
    so they were still trying to pull together.

    although im not too sure how this distortion would effect the result.
    maybe it could be compensated out.
    this would rule out using an IF filter, wich im using a std ceramic 455khz.
    A dsp to do the IF would be an idea,
    but if I cld get the two frequencies easily and cheaply it would be nice.
    the surest way so far would be to use a DDS and PLL for each,
    but theyre a bit expensive and power hungry for fast ones,
    and does seem a bit of overkill.

    I initialy used just ~100mhz from 2 DDS wich worked well,
    but trying to scale this up to 2ghz was dificult
    I tried multipliers but this proved quite dificult.

    maybe using a DSP to process the signal would allow me to use a lower
    frequency.
    but I think I might try this ramp idea first maybe simulate it,
    should be able to do the comparison at least 100mhz wich should make any
    lock in problems easy to deal with
    I think I could make 4 PDs 1 for each quadrant in a gate array.

    If I could use say 10ghz I would only need to ramp it over 1ghz wich would
    only be 10%
    this would make things easy, but 10ghz laser modulation
    is probably way beyond what I can attain practically,
    at the moment im using a cheap $2 laser pointer diode,
    I was surprised it worked at 1ghz
    2ghz is probaly pushing it a bit realy its droped off quite a bit,
    but it does give a good wide change in phase from small distances.

    the rest of the system works fairly well,
    thanks to various help from this forum,
    especialy to whoever sugested using half a torch reflector
    to collect the reflected light to the detector wich is put sideways on,
    and allow the laser to be in the same path by pasing through the hole in the
    center.

    but its only something I do in spare time,
    so I tend to leave it for ages once I come accros a sticky problem,
    until I think of a neat solution like maybe this one.

    Colin =^.^=
     
  11. Jim Thompson

    Jim Thompson Guest

    You reload synchronously with the counter roll-over. Thus avoiding a
    big step in the edge.

    ...Jim Thompson
     
  12. Chris Jones

    Chris Jones Guest

    If you change the divide ratio whilst the PLL is running, there is quite a
    risk that it will go out of lock for a while. If the reference comes from
    a DDS then it can be swept in a way that makes it phase coherent and the
    PLL stays locked the whole time.

    Chris
     
  13. Phil Hobbs

    Phil Hobbs Guest


    Nah, you use a VCO, mix it with the 1-2 GHz reference, low-pass filter
    the resulting IF output, and then run a 1:1 PLL at the offset frequency.
    You can use a frequency-phase detector at the offset frequency. By
    changing the sign of the loop gain, you can make a USB/LSB switch.
    (Search back for a discussion between Jim Thompson and me on this very
    subject.) Offset PLLs are actually really nice, and no dividers are
    required.

    Cheers,

    Phil Hobbs
     
  14. Chris Jones

    Chris Jones Guest

    Right, that's what every sigma-delta frac-N synth does. I am not sure how
    practical it would be on the common pre-fabricated int-N PLL ICs where the
    frequency is shifted in serially. Hopefully they latch it in a way that
    the timing can be made to work. Certainly in an ASIC or FPGA it could be
    made to work.

    Chris
     
  15. colin

    colin Guest

    Ive tried this, the problem is the loop bandwidth is necessarily
    less than the offset frequency,
    it needs to have reasonable gain at the offset frequency
    so that it can overcome the 2 VCOs from pulling eachother into lock.

    Colin =^.^=
     
  16. Phil Hobbs

    Phil Hobbs Guest

    If you have enough stray coupling to overcome the action of a
    frequency-phase detector, man do you ever have layout problems. With a
    double-balanced mixer, I might almost believe it, but no way can you
    argue with a 4046.


    The topology I'm talking about is this one (copied from our previous
    go-round here):

    USB/LSB----------------------------------------------+
    |
    |
    |
    125 kHz ref --------------------------------+ |
    | |
    _V___ _V__
    8 MHz carrier-------\\---\ RC=700 ns | | | X |
    )) >---RRRR---+--->| PFD |---| +-1|---------+
    +-//---/ | |_____| |____| |
    | R
    | (xor) CCC R
    | | +-CCC--RRR--+ R
    | GND ___ | /| | |
    | / \ | /-|----+---+
    +-----------<----------| VCO |<--+---< |
    | | \+|----+
    \___/ \| |
    GND

    For your situation, you'll need a 2 GHz DBM instead of an XOR, but the
    idea is the same.

    Cheers,

    Phil Hobbs
     
  17. colin

    colin Guest

    I made such a prototype with 2 vco modulules, mixer, and 4046 on a small
    board,
    it would pull so bad it wouldnt lock at anything less than 10mhz offset,
    I moved the vco to 2 seperate boards, with sheilding, triple buffers etc
    it worked ok at a few mhz but although it will lock at 500khz
    the signal from the mixer is very distorted.

    im looking for a solution where I can put everything on the same board
    again.

    both vcos are amplified and fed to optical devices wich makes it difficult
    to shield totaly.

    I think if I have an xor type PD and a flip flop wich detects wich vco is
    ahead
    and steers the xor output into pos/neg charge pumps
    that will give me a phase error ramp wich covers 360'
    obviously il need to prescale the input a bit, cplds seem to be able to cope
    with ~500mhz.
    wich still allows me plenty of loop gain at many Mhz.

    it would be best to avoid the discontinuities in the ramp so inverting the
    signal
    will allow me to keep to +/- 90'

    Colin =^.^=
     
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day

-