Optimizing The Ribbon Transformer Design
I've done some reading on the web about how to reduce parasitic transformer losses. With the relatively small number of turns on a Ribbon matching transformer, it appears that capacitance between turns is not important - it's a different story if you a build a vacuum tube output transformer with thousands of turns. Wire resistance losses could be significant (because of the high secondary currents) but are handled by using big wire gauge in the secondary.
The most significant loss for my application is leakage inductance. You can measure it by shorting the secondary and measure the inductance across the primary. This leakage inductance can limit high frequency response and also can cause high frequency phase changes. This can be reduced a lot by using bifilar, trifilar, .... - generically called multifilar winding. For bifilar winding, you take two pieces of wire, twirl them together (not tightly) and then you wrap the pair around the toroid. Since you have two wires, you wrap the pair half the number of turns you've calculated (times two wires for each turn gives the correct number of turns). You have to connect the two primary wires in series in order to have the right number of turns. It's hard to describe in words so I've attached a diagram to show how it is wired for a three wire case (trifilar). This gives you a more efficient coil. It's not intuitive to me why it works but I’ve read it was first invented by Tesla (yes, THAT Tesla)
In the test I've done, I compared a standard wired toroid (one wire primary with a single wire secondary wrapped over it) to a toroid wired per the picture. They both had the equivalence of 30 primary turns and 5 secondary turns. I wired the primary in series trifilar. I think this has the most effect on leakage inductance. I wired the secondary with three wires in parallel. I think this mostly helps current carrying although it also has an affect on leakage. (I've not seen the terms "series" and "parallel" applied to these windings but it makes sense to me so I'm going to use it here).
The bottom line; The bifilar wound transformer had leakage inductance about 1/3 of the standard windings. The standard transformer had leakage inductance ~ .5 mh and the bifilar was ~.16 mh. The primary inductance was slightly higher for the bifilar transformer (~5-10%) and the secondary inductance was unchanged. I also tried as high as hexfilar (6) primary windings with the primary and secondary twirled together (for maximum coupling). This made very little improvement (probably within my measurement accuracy).
Regarding transformers; BG Micro also has a good price ($2.00) on what I expect to be excellent toroids for the matching transformers:
http://www.bgmicro.com/prodinfo.asp...&time_out=44:50
It has a little bigger cross section than the ones I used on the prototype. It is made of type "J" ferrite which appears to be better than the type "77" in my prototype toroids. The BH curve of the "J" is more linear than the "77" (the 77 has an odd wiggle that I have not seen on any other ferrite). Also, the two parts of the curve (backwards and forwards) are much closer and quite linear. I think this will reduce core losses and distortion.
I've ytied winding these cores with several different arrangements to minimize leakage inductance. I ended up with a primary consisting of three series (per my previous post) multifilar windings of 17 turn each, using 18 AWG wire. This is the equivalent of a single 51 turn primary. For the secondary I used twelve parallel windings of # 24 AWG with nine turns. This gives it roughly the current carrying capacity of 13 AWG. The parallel secondary windings cut the leakage inductance by a factor of 2 or more over a single secondary. The combination primary and secondary multifilar windings I used ended up with a leakage inductance of .09 mh, measured with the inductance range of my multi-meter. Using a single 57 turn primary and a single 9 turn secondary gave me a leakage inductance of 1 mh. This is an excellent reduction.
Note that my multimeter uses a fixed frequency to measure - probably something like 1k. The leakage inductance is effectively in series with the primary so the main concern is its value at 20k. I did a simple test of putting a four ohm resistor in series with the primary while I had the secondary tied together (this is how you measure leakage). I then ran 20 khz through the pair and found (by ratio of voltage drop) that the XL of leakage inductance at 20k is ~.8 ohms. This works out to .0064mh at 20k (if my math is right). This is quite good and means the high frequency loss due to leakage inductance (assuming an equivalent 8 ohm primary load) is much less than 1 db.
I've included some pictures showing the steps in building the transformer. It's the same design as the first:
Primary = 18 awg, 17 turns, 3 wire series multifilar winding (equivalent to 51 single wire turns)
Secondary = 24awg, 9 turns, 12 wire parallel mulitifilar.
The top left picture shows the primary with the three wires per turn.
The top right picture shows how the three wires are connected to get them series connected. Note that it's a good idea to mark each end of the three wires to make it easier to figure out what to connect.
The third picture shows the secondary wound on top of the primary. No need to keep track of individual secondary wires since they are all wired in parallel.
The last picture shows the completed transformer wrapped in electrical tape to keep the windings tight to the core.
Transformer Response Measurement
I measured the transformer response by connecting to the secondary of the transformer instead of the usual mike input, and keeping the JustMLS "probe" on the primary of the transformer. Then I used the subtract function to isolate the transfer function of the Xformer. I've attached the results and the response is pretty close to dead flat, within the limits of the sample rate and window size. That's about all I can measure until I get software that can measure distortion. I think you can ignore the phase part of the plot. It is a results of not correctly setting the offset when measuring. I want to get a soundcard that can handle sample rates higher than 48k - although it is of only theoretical interest since I can't hear much above 15k.
