Re: Two series spiral inductors

Yes. The spirals are coiled up TRL, but as long as it is electrically small compared to wavelength. It still can be treated as lumped choke. This is a valid case for frequencies at a few GHz and below. It 's the parasitic capacitances that mess up the inductor bandwidth. Do you agree?

What does the miniature axial leaded chokes look like? Is it just a regular air-core inductor using the gauge wire? As you said, it sits above PCB and behaves better. --another evidence that parasitic capacitance is the actual culprit, not the TRL effect.

I agree that lumped choke has a worse ESR that can hide the resonance problem to certain degree. You are a gentleman.

---------- Post added at 22:22 ---------- Previous post was at 21:56 ----------

Great comments!

---------- Post added at 22:25 ---------- Previous post was at 22:22 ----------

We all lucky dudes met here and sparring with same interest.

Agreed. I've seen designs and papers where there is series L, shunt C, series L etc. These need careful design to reduce resonances as you say. I'm perhaps a 'poor customer' for this type of wideband choke because small suckouts are not good news for many of my applications :)

From experience, adding damping resistors can help reduce the sharpness of suckouts but I have often found that they spoil the through loss on a bias tee. Eg adding a few tens of ohms in series with a bias tee choke gives a lower shunt impedance than just the bias tee choke on its own.


I did take one of these chokes apart a few years ago and it is nothing more than a tiny axial inductor wound on a tiny former. I'm pretty sure the former was a grade of ferrite.

As for the spiral inductor I can't see how you can build up 1uH without getting electrically long at your upper frequencies.

I guess you could try making a few in series but I think there will be resonances to spoil the performance.

I guess that is part of the fun of RF though. Sometimes you can find new RF techniques at the same time as you prove or fail the original approach :)

Wonderful comments!

One thing intrigued me is: for the two spiral in series case (i.e. 100nH+1uH), is the resonance caused by the small cap from the 1uH resonating with the 100nH, or is it caused by the small cap from the 100nH resonating with the large L of the 1uH? Maybe both.

I really think that the performance will be poor. I don't have a lot of experience of using spiral inductors for wideband applications so I can only guess as to what the makeup of the resonance will be. I have always considered them to be narrowband eg for oscillator design or perhaps a filter.

Maybe someone else can chip in here but I think by the time you get 1uH out of a spiral inductor down at VHF it will already look bad at the lower end of UHF let alone >1GHz. This is just for one spiral.

If the spec is relaxed to >1GHz then I still think that the moment you add the second spiral there will be a bad resonance.

I think this series technique has potential problems with lumped inductors but it will be much worse for spiral inductors. Also, they are not ideal for arranging in series as you need to connect to the centre for the next series node.

For the commercial bias T, how are they doing in bandwidth? Do they encounter the same problems as we discussed above?

---------- Post added at 21:07 ---------- Previous post was at 19:37 ----------

an interesting finding.. read on.
thought we worked through this thing, and nailed down the multi-spiral in series problem. However, something from Picosecond lab for the wideband bias T looks interesting:

"Manufacturers making narrow, octave-band bias Tees can get away with usig only a single inductor optimized for that band. One cannot use a single inductor for broadband bias tees. Thus for broadband bias Tees, several inductors are connected in series. Each inductor is optimized to cover various frequency bands from microwave, UHF, VHF, HF, and MF down to audio frequencies..."

The above is written by the founder from Picosecond lab is a IEEE fellow. They have ultra-wideband product for this as well. So they must use a special network technique to get away the multi-spiral in series problem....

I really don't think it will be as simple as putting a whole load of inductors in series. I would guess that a bias tee to work from audio frequency to the microwave band would be a very complex network.
It won't just be a string of inductors in series (for the reasons already given in the thread)

Maybe they are using the series L, shuntC, series L, shunt C approach (I asked back in post #10 if this method was 'allowed' in your challenge) . Also there may be resistance included in the network. Also I would suspect such a bias tee won't be suitable for handling much RF power.

Send him an email and ask him.

I reckon he will confirm that the bias tee is NOT just a suite of inductors in series despite what the text above says. It will be a more complex network than this.

What would also be interesting to see would be what DC blocking cap (if any) is fitted on the input side of this bias tee. it would have to produce a low reactance over a very broad bandwidth as well (without resonances of its own)

You are right. The writing was blurry for obvious reasons. He admitted R, L, C are used, and it is a complex network. Not just as he stated above "inductor in series". Strictly speaking, he cannot take it for granted by saying that. As a matter of fact, in his note, he used R and C to controll Q (BW) to smooth out the cross-over band.

Shouldn't the shunt caps make it worse?

I think the idea is to make the network lossy. eg trade some thru insertion loss to get bias tee resonances that are damped. So maybe they fit resistors (and maybe inductance) in series with the shunt capacitors.

I would imagine that this is one of those tasks that you just get better at the more you experiment.

Each time you find a new 'contender' for one of the series chokes you probably have to optimise everything again to achieve the next level in performance.

Pico have probably been tinkering with these complex networks for a long time.

Edit: for a bit of fun I quickly simulated and knocked up a wideband bias tee using series and shunt components and managed to get -0.25dB loss or better across 300kHz to 3GHz. This matched the simulation.

This is the max end to end range of my VNA. I reckon I might be able to get to 10GHz with another section (or two) but I can't measure this at home :(

That is great! Could you please post the schematic you used for this 300K to 3GHz wideband bias T?

Hi
I brought the circuit into work today to look on a 6GHz VNA and it holds out pretty well to over 4GHz. Actually my test fixture is the weak link so I'm going to make a decent test fixture and try again.

I optimised the circuit on a linear simulator. The resistor values in the shunt circuit came out quite large!

Here is a basic circuit although my circuit on the linear simulator had extra stuff to model the PCB and parasitics etc.

tee2.doc

I'm lucky in that I have a T-Tech 7000 PCB milling facility here at home so I might just make up a decent version of this on Rogers 4003 material.

The only downside is that the larger series chokes can only take 110mA max current. Therefore I'd like to swap out for something better. However, even if I manage this, the bias tee is only suitable for low RF power due to the (damped) suckouts that are still visible but only a fraction of a dB. Also there is some inevitable resistive loss.

I also tested a Mini Circuits bias tee at work. This worked to 6GHz but gave over 1.2dB loss by 6GHz so I'd like to see if I can better this :)

http://www.minicircuits.com/pdfs/ZFBT-6GW-FT+.pdf

What happened to frequency below 300KHz? Is that the low end limit to the VNA you have? I assume that your bias T should work down to DC, even if it is not measureable, since the schematic you attached is indeed a low-pass filter configuration.

The resistors are added to shunt C to reduce its Q, and to damp the resonances (as you indicated). The placement of shunt RC actually broke the "series L only" network, and improved the bias T bandwidth. This improvement may be explained by using a network analysis. I haven't figured out exactly why this change helps to extend the bandwidth in network terms yet.

Pretty cool that you have the luxury to own a home lab with machines to play with! I guess you can make patch antennas too.

Yes, as mentioned if you do add a series resistance somewhere to dampen out the ripples, you will have more insertion loss in the main arm. You want to put the resistance at the far side (DC side) of the inductors if possible, so that the additional loss only occurs at the lower frequencies where you typically do not care so much.

That is a nice trick!

I combined all of your aspects, and simulation RLC with it's spice model, the result is following: it seems almost to reach Mini ZX85-12G performance.

Please a schematic for your simulation results.

Here you are, the schematic