EMRFD Message Archive 11053

Message Date From Subject
11053 2015-05-01 21:40:40 Ashhar Farhan doubly balanced diode mixers and RF-IF port isolation
Attempting to make a bidirectional mixer for the unending Minima project, I needed good IF to RF port suppression. Surprisingly, I read very little of the literature at hand. Most of the specsheets (of ADE-1 and SBL-1) only mentioned the LO-to-RF and LO-to--IF port isolation. I couldn't glean much from EMRFD either. I tried squinting at the screen-shots of erstwhile qrp pops mixer pages: no joy. 

This had to be measured.

I made mixer with two balun cores and four 1N4148 diodes. the construction details are at the fag end of this mail.

Having done that, I placed this mixer in a test jig. It is important to name parts now ...
Transformer A : this is the one with it's center tap grounded.
Transformer B : this is the one that has the center tap designated as the IF port.

Experiment 1.
The standard literature uses the diode mixer with LO fed to transformer A (the one that has the grounded center tap). The IF port on transformer B was connected to a 24 MHz, -10dbm source from an Si570. The RF port was connected to the homebrew spectrum analyzer. The LO was tuned to 45 MHz as I was interested in generating the 21 MHz output (the closest to the IF frequency). The Specan was directly connected to the RF port of the mixer (primary of Transformer B).
1.1 The scan was quite noisy until the LO was increased to around 8 dbm. I used a 100 ohm resistor from the Si570 (CMOS) to a 6db pad to get to that level.
1.2 The scan improved. The output at 21 MHz was at -16dbm, the output at 65 MHz was higher by 2 db! This could be the ripple in the 70 MHz LPF of the specan.
1.3 The LO was leaking through at -50 dbm. The IF was down at -50dbm.

The trouble with this was that, given the wanted signal of -10dbm at 21 MHz and the IF breakthrough at -50dbm, The IF to RF isolation turned out to be just about -40dbm. This was unacceptable level even from a compliance point-of-view.

A closer look at the circuit revealed that the IF is taken from the central tap of transformer B while the RF port is the primary. Hence, it was possible that the IF was leaking into the RF due to the imperfection and capacitive coupling in transformer B.

Experiment 2:
1. The transformer A that had the LO connected to it's primary and the center tap was grounded in experiment #1. Instead, this transformer's center tap was lifted and the IF was connected to the center tap of transformer A.
2. The transformer B had IF connected to the center tap and RF port on the primary. Instead, the center tap as ground now. 
3. The Si570 which is very rich in harmonics was replaced by a more elaborate circuit. A 24 MHz crystal filter with buffer amplifiers at both ends was used to drive the IF port. The Si570 was routed through the crystal filter. This was done to reduce the harmonics of the Si570 and provide the mixer with robust 50 ohms drive. The Si570's output was tuned to provide 100mv peak (-10dbm) on the power meter before connecting it to the mixer.
4. A low-pass filter (4 section) with a cut-off of 22 MHz was inserted between the RF port and the specan.

The scan was absolutely clear! The IF output is now gone down to grass level (-below -75 dbm). The LO is more than -60dbm below. 

Experiment 3:
Encouraged by this, the LO was retuned to 27.5 MHz to provide an RF output at 3.5 MHz. Now, the output at 3.5 MHz was accompanied by a 7 MHz spur as well that was only 30 db down.

The Max RF-to-IF isolation configuration of the diode mixer should use the center tap on the LO tranformer for IF input and the center tap of the RF port transformer should be grounded.

Where the IF-to-RF port isolation is needed, the IF has to be taken from center tap of the transformer that is used to drive the LO energy. The transformer connected to the RF port must have its center tap grounded.
This  "Max RF-to-IF isolation configuration" may not have great LO to IF suppression (which is fine in my case as the IF port is drive to and from the crystal filter in my bidirectional rig).

Foot note on Construction of the diode mixer

I am giving these details as physical construction of a diode mixer can impact the balance. 

1. I set out to select the diodes. First, from a batch of 20 1N4148s, I batched them by equal forward voltage drop when fed by a 2.2k resistor connected to a 12v supply. I found that the forward voltage drifts for 30 seconds before settling down. My cheap DVM measured it on the 2 volt range. I am suspicious of the accuracy of its readings (Precision doesn't mean Accuracy - a gem from the SSD).
However, a bunch of diodes with 0.685 volts was selected. To be sure, I rechecked the batch again choosing this particular batch. Found the readings to be the same.

2. Using my TV baluns cores, I wound two trifilar transformers with SWG #28 wire, wound as best as I could manage, I drew the bundle firmly but not tight and pressed the wiring against the inner walls. Six turns to each transformer, a drop of superglue on each side of the balun to hold down the windings. I kept the inductance lighter than usual as the RF and IF were above 20 MHz and I wanted to keep the capacitive coupling between the windings to a minimum.

3. The mixer was assembled on a small copper clad. in the center i cut a 2x2 pad where the four diodes were assembled in a ring. the transformers were soldered belly up.

- f
11062 2015-05-03 08:18:51 kb1gmx Re: doubly balanced diode mixers and RF-IF port isolation
Things often forgotten with DBMS...

MiniCircuits has a few apnotes on the subject and there are others out on the net
worth a read.

If you use junction diodes (1n914, 1n4148 etal) the drive needs to be higher as
those diodes have a higher junction voltage.  The side effect is they produce a
DBM that can stand higher (about 3-6db) levels depending on transformer 
construction.  The higher levels com form the fact those diodes can stand 75ma.
This can be important to higher signal situation where a "stronger" mixer is 

Commercial DBMs use Schottky (hot carrier) diodes that have a lower threshold
than junctions types.  The direct effect is this lowers the drive need for mixers
and also the conduction curve is different so the impedance is more uniform.

Last item in diodes, whatever is used must eb 100% matched for all parameters
and dual or better monolithic quads usually do best for matching and also 
temperature tracking.  Both are important if only to recognize differences in 
hand made to high performance mixers.

FYI: matching is precision operation requiring no accuracy.  You care not that
the meter read x.xxxx volts only that all matched dies read the same.   Matching 
is best done with a bridge circuit with a calibrated (by hand) balance pot.
Also diodes are temperature sensitive, heat from ones hand will upset match
also current through the diode.  If you running the diode at a known RF power
do the match to get the diode current and use that value as the source current
rather than a random 2.2K from 12 volts (about 5mA).

Another is that the transformer symmetry is very important to isolation and balance.
Some of the commercial transformers use a 1:1 unun from the unbalanced input
to the transformer winding (RF and LO ports).  reason for this is if one end of the 
primaries is grounded there is a slight difference in capacitance to ground for 
that end of the secondary and some imbalance results and reduction of isolation
as well.

You have already observed the IF port lead (which side of the DBM relative to 
the LO) has an impact to the IF LO isolation.  Also how the IF is terminated
has an effect that can be significant.  To that does the termination have a DC 
path to the other end of the mixer (the other center tap)?   Is the terminating 
impedance correct (not always 50 ohms) are the other ports terminated in 
the same impedance?  This is critical to SI570 and other NCOs since they 
are NOT 50 ohms sources. The closest approximation is they are voltages 
sources up to the current limit of the device. Most CMOS sources have two 
additional things to watch one is they often can sink more current than they 
source (unequal impedance in high state VS low)  and the other is driving 
leads (from part to load) may induce transformations depending on frequency.   
So if your driving something that is impedance sensitive attenuating at the 
NCO and then buffering with a low can amp to get a consistent driving 
impedance is a more predictable method.   Isolation also varies with 
driving power and driving impedance at the source and both interact 
with the port impedance that varies with drive.

In the end for best results impedance and levels must be controlled.

With DBMs spur and frequency analysis is best evaluated to not less than 
harmonics and maybe as far as the 11th.  Part of this is square wave drive 
and also because the diodes int he mixer will turn sine waves into square 
waves by clipping.

Mixers are complex circuits due to the transforms and nonlinear behavior.
Also every communications system we use has them so understanding them
is valuable.