EMRFD Message Archive 703
Message Date From Subject 703 2007-05-03 16:19:08 ka3j12 VFO level for SA612 Mixer Hi - I understand that 700 mV pk-pk is about what the SA612 likes to
see when used as a typical first mixer in a superhet. Has anyone
experimented, analyzed or care to speculate regarding the impact of
running with a lower VFO signal level? Thanks!
704 2007-05-03 16:51:03 Luiz Amaral Re: VFO level for SA612 Mixer This is said to be the condition for best conversion gain and best 3rd IMD
Luiz - PY1LL / PY4LC
----- Original Message -----
706 2007-05-03 17:23:59 DSNman@comcast.ne... Re: VFO level for SA612 Mixer Ron,
I would guess it would cause lower conversion gain and maybe more easily overloaded by strong signals.
-------------- Original message ----------------------
From: "ka3j12" <firstname.lastname@example.org>
> Hi - I understand that 700 mV pk-pk is about what the SA612 likes to
> see when used as a typical first mixer in a superhet. Has anyone
> experimented, analyzed or care to speculate regarding the impact of
> running with a lower VFO signal level? Thanks!
> Ron (KA3J)
707 2007-05-03 19:12:08 Kevin Purcell Re: VFO level for SA612 Mixer From the datasheet for the 612
> External L.O. should be 200mVP-P minimum to 300mVP-P maximum.so 700mV is pretty hard drive though they do drive the 602 with 630mV
on EMRFD page 5.10 which is a bit high. Curiously they get an IIP3
of -17.5 dBm which is rather worse than the typical -13dBm (though
their measured gain is also 3dB up which might explain it).
With the 602/612 you don't drive the "top" transistors directly there
is the oscillator transistor, biased for about 0.12mA emitter
current, and an unspecified buffer in the way. But I'm pretty sure
they're in the hard switching regime. It might be that given the
design of the buffer amps it's difficult to get to the linear regime
(if the buffer is not very linear).
You might to read the 1496 app note that I posted a link to a few
days ago as I think it may be relevant here.
Both mixers are gilbert cell designs but the 1496 can be run in
either "switching mode" with high level drive or a "linear mode" low
level drive with the crossover point being 15 to 20mV RMS or 40 to
60mV pkpk (though this varies depending on the gain of the device and
the standing current in the diff pair).
For the 1496 the voltage gain drops to about half in the "low drive"
case but so does the number of spurious outputs (no more odd
harmonics in the LO). The drop in gain should improve the input IP3
and so the spurious free dynamic range (in the 1496 at least). The
noise figure also increases as the gain drops but for HF that might
not be a bad point. Also amplitude variations in the LO will appear
in the output too (but you should be able to mitigate that).
The gain also drops as you increase the diff pair standing current
but NF improves.
The original Gilbert patent 3689752
and this one is interesting because it has an expression for the IMD
and how it depends on the emitter current (more is better), emitter
resistor (larger is better) and the input impedance (lower is better).
IP3 = 10 + 20 log(104) + 30 log (1 + (R_E I_EE)/104) - 10 log (R_i/50)
R_E is the emitter degeneration resistor in kohms and I_EE is the
emitter current in mA. R_i is the input impedance in ohms.
By inspection the IP3 goes as the cube of R_E and the cube of I_EE.
So turning up the current has a large effect on IP3. Likewise turning
down the gain has a strong effect on IP3.
For R_E=1kohm; I_EE=1mA; R_i=1.5kohm the input IP3 is -15dBm. Seems
For R_E=1kohm; I_EE=10mA; R_i=1.5kohm the input IP3 is +15dBm. Seems
good. Perhaps too good. I'm pretty sure these parameters interact
with others :-)
The relates to some earlier comments on 1496 thread by Allison.
I'm still writing a response to that but the idea is it would be
interesting to see the 1496 characterized for high standing current
(~10mA max), low gain and low LO drive (rather than from it's normal
parameters -- 1mA standing current and high level drive) and see
what effect that has on the intercept and noise figure. The part is
also designed to run at up to 30V. I wonder what effect increasing
the Vcc has on IP3?
Fo example the discrete Gilbert Cell mixer on p5.11 was biased at 5mA
and 15mA and gives much better numbers (as you might expect) with an
SSB noise figure of 13.8dB.
15mA standing current, gain 18dB, IIP3 +11dBm with 0dBm LO drive
5mA standing current, gain 16dB, IIP3 -2dBm with 0dBm LO drive
So given that and the other expression presented above it seems like
it might be worthwhile to characterize these devices when "pushed".
They're not going to exceed a diode ring mixer but they might still
give decent results (and be easier to get hold of in VU-land).
The other cryptic comment (one of the many!) in EMRFD is:
> Although the main tool used to improve IMD performance in a GilbertAnd that's Chris Trask, N7ZWY, but unfortunately it's not an online
> Cell is to increase current, feedback can also be applied. The
> experimenter should examine the work of Trask.
paper but some of his other active mixers with lossless feedback
After all EMRFD is about experimentation, right.
709 2007-05-04 11:02:23 Dave Re: VFO level for SA612 Mixer 712 2007-05-04 13:33:54 Kevin Purcell Re: VFO level for SA612 Mixer Did you measure any other parameters?
717 2007-05-04 18:50:46 Kevin Purcell Gilbert Cell mixers with good performance Of Chris Trask's papers this one is perhaps the most relevant to this
discussion -- comparing a linearized Gilbert cell mixer with feedback
to a diode ring mixer. It's nice overview of trade offs with
different mixer designs too. And he reports measured numbers!
Notice the "regular" Gilbert cell mixer parameters with small emitter
degeneration resistor and biasing set to keep the lower transistor
linear at all times. With a 15mA standing current and 0dBm LO drive
P 1dB +4.5dB
No mention of the noise figure.
> It has long been recognized that the mostand then goes on to show his series/shunt feedback amplifier
> serious limitation in the IMD performance of tree
> mixers is that of the voltage-to-current conversion
> of the driver transistors Q3 and Q6 Various
> methods have been utilized sucessfully to correct
> this deficiency but these methods all ignore
> secondary sources of intermodulation, primarily the
> hfe nonlinearity of the driver transistors and the
> nonlinear charcteristics of the four switching
> transistors. These deficiencies can be overcome
> by making use of a simple series/shunt feedback
> amplifier circuit wherein all of the transistors are
> embedded within the feedback topology.
modifications. The first version just adds four resistors from the
collectors at the top of the mixer tree to the corresponding bases on
the "same side" of the tree.
With a 15mA standing current and 0dBm LO drive he measures
P 1dB +5.5dB
This technique might work with the 1496 too. You can get to all the
terminals of the transistors to add the feedback resistors (a win
over the 602 for an experimenter) but you can't control the emitter
resistors at the bottom of the tree (though you can control the bias
on the constant current transistors. If I'm wrong on any of this let
me know :-)
The Gilbert cell mixers also have an advantage in being a lot less
sensitive to IF port termination unlike commutative mixers. Below the
comparison is between a perfect termination (Without IF Filter) and
into a resonant bandpass filter (With IF Filter).
Without IF Filter P 1dB −4.5dBm IIP3 +19.0dBm
With IF Filter P 1dB −7.5dBm IIP3 +7.5dBm
Transistor Tree Mixer (aka Gilbert Cell)
Without IF Filter P 1dB +4.5dBm IIP3 +17.5dBm
With IF Filter P 1dB +4.5dBm IIP3 +16.5dBm
Linearized Feedback Mixer
Without IF Filter P 1dB +5.5dBm IIP3 +21.5dBm
With IF Filter P 1dB +5.5dBm IIP3 +20.75dBm
I'm also reminded (being a G and avid reader of Technical Topics in
my youth) of the, now obsolete, Plessy SL6440. It's use was
restricted to mostly British and Commonwealth (G, and some VK, ZL,
ZS) RF designs though it appeared in a lot of commercial and military
radios. It's interesting how parochialism (and availability of
products) constrain designs even when an excellent part is available.
But you can at least still get the datasheet
It was good for HF and low VHF use (3dB down at 100MHz min spec) with
very good numbers with a 250mV rms LO drive (into 1.5kohms)
IIP3 +30dBm (with a standing current of 50mA)
P 1dB +15dBm
Not bad for a Gilbert Cell mixer and even the NF was OKish :-)
A Plessey application note AN-156-2 "Intermodulation, Phase Noise and
Dynamic Range" explains why Gilbert cell mixers have these issues.
it's also interesting for linking phase noise to dynamic range
consideration (and has a fun appendix on where IMD arises and how to
> Many advantages accrue to the choice of the transistor treeThe SL6440 was an improvement on the standard Gilbert cell mixer
> type of approach. Here the input signal produces a
> current in the collectors of the lower transistors and this current
> is commutated by the upper set of switching transistors.
> Because the current is to a first order approximation independent
> of collector voltage, the transistor tree does not exhibit the
> sensitivity to load impedance that the diode ring does, and
> indeed, by the use of suitable load impedances, gain may be
> achieved. The non- linearity of the voltage to current conversion
> in the base emitter junctions of the bottom transistors is the
> major cause of intermodulation, but by using suitably large
> transistors and
> emitter degeneration, very high performances ( +32dBm input
> intercept) can be achieved.
> In an attempt to improve this trade-off, P. E. Chadwick, "TheThat's Peter Chadwick, G3RZP.
> SL6440 High Performance Integrated Circuit Mixer", Plessey
> Semiconductors Limited, 1981, discloses that complementary
> linearising transistors can be added to the lower pair of
> transistors to provide negative feedback. However, this approach
> has the disadvantages of introducing noise and presenting bias
> stability problems for practical ranges of operating temperature
> and integrated circuit process parameters.
BTW he has this to say about low LO drive into a Gilbert cell mixer
> Bill said:So my previous suggestion of trying lower drive and operating the
>> But if the local oscillator signal is strong enough to simply
>> turn on
>> and off the mixer tube it does become a very effective >frequency
>> It then appears to be an analog mixer multiplying a signal with a
> It does reduce the conversion gain, though. That's the big problem
> switching mixers. The so called 'Gilbert cell' (which Barrie Gilbert
> doesn't claim as his - it was first patented by a guy from RCA as a 4
> quadrant multiplier - Barrie's claim is that he was the first to
> using it as a mixer, and even that's disputed) can be oeprated in
> gain as a switching mixer, and in higher gain as an analogue
> type mixer, with lower LO drive and worse IMD performance. It also
> is more
> susceptible to AM on the LO.
> Peter G3RZP.
1496 as a linear multiplying mixer is not a good idea.
So how to compare mixers. That's the next email.
768 2007-05-14 20:52:37 Kevin Purcell Re: VFO level for SA612 Mixer Whilst hunting around for other information I came across Cosy MUTO,
JH5ESM, who has measured this for a 602 + NJM4580 op amp.
> Gain is proportional to LO levels under -20[dBV] (0.1[Vrms]).．He gives gains in dB but as they're not between matched impedances
they're not really power gains just "voltage gains converted to dB".
> When the gain is proportional to LO level (i.e., under -20[dBV]),He concluded for a DC RX front end
> AM feedthrough seems to be constant. However, it goes wrong by 10
> [dB] when the mixer is overdriven.
> Therefore, LO level should be carefully chosen to obtain best
> results, that is -20[dBV] in this case.
> Optimum LO level is -20[dBV] (=0.1[Vrms]) under 5.0[V] supply voltageWhich is about 280mV pk-pk.
777 2007-05-20 13:45:11 Dave Re: VFO level for SA612 Mixer Kevin,
Sorry for the delayed reply...I don't check the Yahoogroups very
often. No, I did no other measurements