Connect with us

Discrete low noise wideband transimpedance amp

Discussion in 'Electronic Design' started by Martin Devera, Feb 16, 2004.

Scroll to continue with content
  1. Hello,

    I'm trying to make LNA preamp for photodiode with BW 100MHz. I was
    interested in lowest noise I can get. I tried both JFET and BJT
    circuits and here is the best one till now:

    http://luxik.cdi.cz/~devik/tmp/diffpd.jpg

    It is classical cascoded diff-amp with BFR93 (readily available
    here for $0.20 in 1pcs quantity). It is almost flat to 80MHz with
    -3db at 115MHz with transimpedance gain 10kohm (which is my goal
    because input signal is 100nA..100uA AC).
    It has input noise 25nA over 100MHz BW (2.5pA/sqrt(Hz)). PD is
    modeled as one with 3pF with given bias. R6=200, R4=R5=1k, others
    are not important.
    The design seems stable, only part where one needs to be careful
    is R9=R10=10k which is sensitive to paralel capacitance (must be
    under 0.1pF).
    Most of noise is from base shot noises.

    Design with simple cascode with the same feedback as in above had
    35nA of noise. However TZA3033 has 16nA over 90MHz BW.
    I'm interested how should I design amp with yet lower noise :)

    Any suggestions ?

    thanks, Martin
     
  2. Brian Howie

    Brian Howie Guest

    The usually topology for these types of circuit is a GaAsFET with PNP
    Cascode plus emitter follower . The feedback resistor is taken between the
    emitter follower output and the gate of the GaAsFET. You can get down to
    1pA/rtHz or better. Keeping the capacitance across the f/b resistor is
    tricky in discrete -maybe a couple in series would work.

    I've a circuit, but it's proprietry - however there is one here that's
    similar. I don't like the look of the inductor in the feedback on that one.

    http://www.discovercircuits.com/PDF-FILES/fetamp0.pdf


    Brian

    --
    Brian Howie
    BAE SYSTEMS Avionics Limited
    Sensor Systems Division
    Crewe Toll Phase II, 1st Floor,
    Edinburgh EH5 2XS
    Phone +44 (0)131 343 8769
    FAX +44 (0)131 343 8941
    Email
     
  3. The circuit shown does not appear to be biased very well. It seems to be
    relying of the base current drop through R7/R8, to drop the base voltage
    so that there is some Vce available at Q3/Q1. Typically base bias
    resister are setup so that their volt drop is negligible, otherwise bias
    voltages can be highly dependant on hfe and temperature. Tt 1ma, hfe=50,
    the drop is 10V. If you don't have any drop the transistor will
    saturate.

    I ran it is SuperSpice:), with a typical set of worst case values. The
    collector current ranged from around 1ma to 4ma, with a corresponding
    3db variance of 66 MHz to 300Mhz. This was a quick test, with models I
    had to hand. Midband noise for all cases was as you claimed at around
    2.5pa/rthz. Max and min noise at 100Mhz was 4.2pa to 2.8pa.

    Kevin Aylward

    http://www.anasoft.co.uk
    SuperSpice, a very affordable Mixed-Mode
    Windows Simulator with Schematic Capture,
    Waveform Display, FFT's and Filter Design.

    "That which is mostly observed, is that which replicates the most"
    http://www.anasoft.co.uk/replicators/index.html

    "quotes with no meaning, are meaningless" - Kevin Aylward.
     
  4. Ok, second go, I replaced the diff pairs with ATF551 fets, and added
    resistors from gate to the negative rail of 200k.

    This got me noise of 1.1pa out to 10Mhz, raising to 1.5pa at 100Mhz.
    Current was about 7.5ma, 3db down at 350Mhz.

    ..MODEL ATF551_XN nmf(Vto=300m Beta=444m Lambda=72m Alpha=13 Fc=650m
    Cgs=619.3f
    + Cgd=143.5f Rd=2.025
    + Rs=675m Is=0.1f KF=1.e-003f Betatce=-500m Vtotc=-2.5m AF=1)


    Kevin Aylward

    http://www.anasoft.co.uk
    SuperSpice, a very affordable Mixed-Mode
    Windows Simulator with Schematic Capture,
    Waveform Display, FFT's and Filter Design.

    "That which is mostly observed, is that which replicates the most"
    http://www.anasoft.co.uk/replicators/index.html

    "quotes with no meaning, are meaningless" - Kevin Aylward.
     
  5. The design seems stable, only part where one needs to be careful
    Yes, you are right. I did pay no attention to biasing, I wanted to get decent
    noise properties vs. BW first. Also I'm not sure whether it is good to use
    diffpair from discrete tranies given that I can't buy matched UHF parts here ...
    In any case thanks for your comments.
    Martin
     
  6. Ok, second go, I replaced the diff pairs with ATF551 fets, and added
    Wow. ATF551 ? I tried to look it at google (for DS) but no success :-|
    Do you have DS ? If 444m is huge transconductance (I never worked with
    mesfets so ....).
    Did you replace all four Q by fets or did you removed whole cascoding ?
    I'm looking at Brian's idea just now. The example PDF gives me about
    1.5pA/rtHz which is very good given that it uses ordinary jfet and
    switching PNP. Only I can't get CB stage past 20MHz yet.

    Martin
     
  7. Oh? I usually make sure that the basic biasing is reasonable. Biasing
    can sometimes be a bit of a problem.
    You dont really need matched parts.
    Did you check out my other post on the ATF551? I did a bit of playing.
    Even with RF=20k, you can still get 150Mhz at 3db. The max noise is then
    about 1.2pa, with 0.8pa up to 10Mhz. Total integrated noise at 100Mhz is
    9na, about 8.8db improvement on the 25na.

    Kevin Aylward

    http://www.anasoft.co.uk
    SuperSpice, a very affordable Mixed-Mode
    Windows Simulator with Schematic Capture,
    Waveform Display, FFT's and Filter Design.

    "That which is mostly observed, is that which replicates the most"
    http://www.anasoft.co.uk/replicators/index.html

    "quotes with no meaning, are meaningless" - Kevin Aylward.
     
  8. Its an Agilent part. An enhancment jfet! Typical on voltage is 0.3V-0.4V
    I tried both. Keeping the bipolar for the cascode appears to be better
    BW wise.
    As I noted in my other post, 0.8pa achievable up to 10Mhz, around 1.2pa
    at 100Mhz.

    I have put the data sheet on my web site

    http://www.anasoft.co.uk/ATF551M4.pdf

    Grab it while you can, as I will remove it later. I can email my SS
    schematic if you like.

    Kevin Aylward

    http://www.anasoft.co.uk
    SuperSpice, a very affordable Mixed-Mode
    Windows Simulator with Schematic Capture,
    Waveform Display, FFT's and Filter Design.

    "That which is mostly observed, is that which replicates the most"
    http://www.anasoft.co.uk/replicators/index.html

    "quotes with no meaning, are meaningless" - Kevin Aylward.
     
  9. Did you check out my other post on the ATF551? I did a bit of playing.
    Yes I did (other post). I changed parts in Brian's PDF (thanks Brian!)
    and pasted the result below. It uses non-wideband parts now (BF550,
    BF840, BF861) and has noise (integrated 100Hz-50MHz) 1.8pA/rtHz. However
    its -3db is only at 60MHz.
    I'm not circuit expert so I have to look why CB stage toss most of AC current
    over 20MHz into base instead of into collector (even when I used 4GHz PNP).
    I remember I was able to do 1.2pA/rthz with it but can't remember parameters.
    It is already good for me (I need cheap way up to 50MHz because I'll decode
    it with NE564 which is 50MHz too) but it is interesting challenge where we
    can go :)

    Martin

    ..model dx d(is=1e-15)

    ..SUBCKT photo 1 3
    d1 1 2 dx
    cj 1 4 5p
    vc 4 2 0
    rj 1 5 1Meg
    vj 5 2 0
    rs 2 3 10
    is 2 1 dc 1u ac 1u sin(0 1n 100Meg)
    ..ENDS

    vccp 10 0 5

    x1 3 10 photo

    vd 10 9 0
    rd 5 9 300
    j1 5 3 4 BF861B
    rs 4 0 500
    cs 4 0 1u
    rg 3 10 100k

    vi1 5 6 0
    q1 16 12 6 BF550
    vi2 15 16 0
    c11 12 0 100n
    r11 12 0 10k
    r12 12 10 5k
    r13 15 0 2k

    q2 10 15 25 QBF840
    r21 25 0 1300

    rf 25 3 10k
     
  10. Thanks I got it. Email me schematic if you can (I use 3f5 not SS). By the
    way are these HEMT parts available to amateurs in few pcs quantity ? And
    what price ?
    I can buy whole amplifier with AGC for $8/pcs (5pcs) so that if I'll go discrete
    it must be under $2 (whole 10..20kohm preamp)..

    Martin
     
  11. Guy Macon

    Guy Macon Guest

    I have been following this thread with great interest, so if you
    get schematics, part prices, comments, etc. by email, putting
    them somewhere accessable on the net and posting a URL would
    be most appreciated. If you don't have web space, you can use
    some of mine.
     
  12. gwhite

    gwhite Guest


    That's nice, but why not just go to the source?

    http://literature.agilent.com/litweb/pdf/5988-9006EN.pdf


    more general pages:
    http://www.semiconductor.agilent.co...bemgcgjcfijdio.0&prodCategory=FETs+and+PHEMTs

    http://www.semiconductor.agilent.co...jibemgcgjcfijdio.0&Section=Wireless&oid=78403


    Cool part by the way.
     
  13. The ATF55143 and ATF54143 were probably 'impossible' parts 10yrs
    ago. The 54143 is interesting especially because of it's incredibly
    good noise match to 50ohms (at UHF), at the cost of fairly high
    supply current. When using a small amount of inductive source
    feedback (for 500MHz use), and considering the positive gate bias
    voltage (not alot different, voltage wise from a Si BJT base-emitter),
    it is almost a dream come true for the 500-3000MHz range. The 55143
    has a little higher ideal input impedance, with typical specs
    suggesting slightly lower drain current and lower IP3 than
    the 54143. (My first choice for a normal 50/75 ohm input/output
    preamp, and no current drain limitations, would probably be
    the 54143 instead of 55143 -- but both are pretty damned good.)

    The 54143 isn't a typical small geometry 10GHz FET that can
    be forced into a low UHF application (with the attendent matching
    issues.) The 54143 has many advantages of the SHF FETs when
    used at UHF, while being a truly better choice for the
    application.

    At $1.70 each, even if one has to purchase a handful of the
    transistors, they won't likely empty a bank account. Considering
    that a BJT that could reasonably function at all in the 1GHz range
    in the early 1980s would have been much more expensive than $1.70
    (I forget how much, perhaps $5-$10?), the $1.70 is a bargain for a
    component that can supply almost 100mw of RF power, while also
    providing a circuit NF of .3-.5dB (internally can probably do
    0.1dB), IN A SIMPLE WIDEBAND 50/75 ohm CIRCUIT!!!, and a hobby
    level hacked together circuit can be biased using feedback from
    the drain with a voltage divider to the gate.

    John
     
  14. The ATF55143 and ATF54143 were probably 'impossible' parts 10yrs
    Just for interest, are there other similar HEMT parts from other
    manufacturers ? Just to can select if Agilent part will be inaccessible
    here in smaller ammounts ...

    thanks, Martin
     
  15. Brian Howie

    Brian Howie Guest

    I tried the ATF551 (snorty device, I salted away the data sheet !!) in my
    circuit and stretched the design to 100MHz ( original was 50MHz) ; basically
    took RF down from 70K to 40K.

    I get the noise about 0.9pA/rtHz up to 10MHz but goes up to 5.5pA/rtHz at
    100MHz . Are you sure you put in the detector capacitance ? :p Noise is
    dominated by the the voltge noise coupled via the detector C in my design ;
    this was not significant in the original application.

    Brian

    --
    Brian Howie
    BAE SYSTEMS Avionics Limited
    Sensor Systems Division
    Crewe Toll Phase II, 1st Floor,
    Edinburgh EH5 2XS
    Phone +44 (0)131 343 8769
    FAX +44 (0)131 343 8941
    Email
     
  16. Yep. I used 3pf accross a current source. I am using a simple current
    source for the diode, with 100Meg accross it. I have a bleed of 100Meg
    to gnd to stop the node floating. I also have 1Meg from the base of the
    cascodes to the fet gates, with 400k to the negative rail. There may be
    some blemish somewhere causing the decrepancy.

    It might be possible to remove all the dc blockers and use the fact that
    there is about 0.4 vgs and make a vbe multiplier with the feedback
    resistor without effecting the noise much.
    A "good" design should be limited by the inherent noise of the devices
    used, not the feedback/bias resisters.

    Kevin Aylward

    http://www.anasoft.co.uk
    SuperSpice, a very affordable Mixed-Mode
    Windows Simulator with Schematic Capture,
    Waveform Display, FFT's and Filter Design.

    "That which is mostly observed, is that which replicates the most"
    http://www.anasoft.co.uk/replicators/index.html

    "quotes with no meaning, are meaningless" - Kevin Aylward.
     
  17. Did you check out my other post on the ATF551? I did a bit of playing.
    Kevin used my original cascoded diffpair circuit (I have to check it too).
    By the way I just crafted design from that PDF you sent me and with AFT551
    and BFG25A it goes to 100MHz, Rf=10k, integrated noise 14nA while it is
    0.9pA to 10MHz and 2.1pA at 100MHz :)

    Martin
     
  18. I get the noise about 0.9pA/rtHz up to 10MHz but goes up to 5.5pA/rtHz at
    I forgot - I use detector capacitance 5pF and noise is from:
    FET:24%, Rs of PD: 24% and Rf:50%.
    Martin
     
  19. Yep. I used 3pf accross a current source. I am using a simple current
    Can you post spice netlist please ?
    Martin
     
  20. *********
    QN2 Node24 0 Node30 BFR91A_XN 1
    QN1 Node23 0 Node29 BFR91A_XN 1
    Z2 Node29 Node1 Node10 ATF551_XN 1
    R11 0 Node5 100Meg
    C5 Node1 NodeC5_1 1u
    C4 Node2 Node5 1u
    R10 Node1 0 1Meg
    R9 Node24 Node4 1K
    C3 Node7 Node5 5p
    I1 Node7 Node5 DC 0 AC 1
    C2 Node1 Node7 1u
    R8 Node5 Node7 100Meg
    V2 Node4 0 DC 12
    R7 Node23 Node4 1K
    R6 Node15 Node10 200
    V1 Node15 0 DC -12
    R5 Node2 0 1Meg
    R4 NodeC5_1 Node23 20K
    R3 NodeR3_0 Node24 20K
    C1 Node2 NodeR3_0 1u
    Z1 Node30 Node2 Node10 ATF551_XN 1
    R2 Node15 Node2 400K
    R1 Node15 Node1 400K

    ..MODEL BFR91A_XN NPN(IS=1.32873f BF=102 NF=1.00025 VAF=51.9033
    IKF=8.15511 ISE=13.9029f NE=1.51292 BR=17.6953
    + NR=994.038m VAR=3.28032 IKR=10 ISC=1.04297f NC=1.18993 RB=10 IRB=1u
    RBM=10
    + RE=763.636m RC=9 EG=1.11 XTI=3 CJE=2.03216p VJE=600m MJE=290.076m
    TF=6.5579p
    + XTF=38.9752 VTF=10.9308 ITF=521.078m CJC=1.00353p VJC=340.808m
    MJC=194.223m)

    ..MODEL ATF551_XN nmf(Vto=300m Beta=444m Lambda=72m Alpha=13 Fc=650m
    Cgs=619.3f Cgd=143.5f Rd=2.025
    + Rs=675m Is=0.1f KF=1.e-003f Betatce=-500m Vtotc=-2.5m AF=1)

    ..OPTIONS TEMP=25
    ..OP
    ..AC DEC 50 100 1G
    ..NOISE V(Node23) I1 DEC 50 100 1G

    ********

    Kevin Aylward

    http://www.anasoft.co.uk
    SuperSpice, a very affordable Mixed-Mode
    Windows Simulator with Schematic Capture,
    Waveform Display, FFT's and Filter Design.

    "That which is mostly observed, is that which replicates the most"
    http://www.anasoft.co.uk/replicators/index.html

    "quotes with no meaning, are meaningless" - Kevin Aylward.
     
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

-