I am attempting to design a diode switched bandpass filter as relays are expensive and unreliable. I figured I'd test the switch first in LTspice. This is HF only (max 12MHz) so nothing too complicated. I'm currently here:
Input power is -7dBm from the LO premixer so I don't push the diodes too hard. There will be a broadband post-filter amp to knock this up to 7dBm to go into first IF mixer.
I tried biasing the V1 source with the following voltages and get the following associated loss:
12v (on): -2dBm - fine
0v (off): -35dBm - not great
-12v (off): -50dBm - good
Does this sound reasonable? Are there any better ways with better leakage / isolation?
I am mainly concerned about avoiding a negative rail to get better isolation if possible.
Edit: to note I changed the diode bias to 10mA after doing the above. Better linearity.
I'd consider the 'standard' DPDT PCB relays, they're not expensive and it's only 12MHz so they should be OK, but it seems to me you should only need to reverse bias the diodes to get better isolation and you can do that by grounding the anodes and feeding a +12V supply to the cathodes of the diodes...
Adding C3 and C4 drops leakage further.
I found this document to be very helpful when making diode switches. I have started to add all these useful docs to my blog. You can download it from here:
http://robs-blog.net/Files/AR_Docs/RF_Switches.pdfBack to back diodes on each end of the filter will give -30db of isolation. And thats with using something like a 1n4007, which happens to be the diodes that elecraft use for TX/RX switching in some of their radios. Ummm also google about, Wez Harward also has a Doc from an ARRL article on diode switching filters.
Found it here:https://www.arrl.org/files/file/Technology/tis/info/pdf/9101024.pdf
Cheers,
Rob
If you placed a (bypassed) resistor in series with each of the chokes to ground, you could then arrange to reverse bias the diodes in the off state by a few volts. This will decrease the diode capacitance as it increases the width of the depletion region.
Some power rectifier diodes are actually a PIN structure and they often have useful minority carrier lifetimes for HF.
Placing an extra switching diode to shunt to output to ground is helpful for isolation, especially because the load on this thing will NOT be 50R over much of the frequency range of interest (Filters are, generally, reflective).
This is over complicated and could be simplified quite a bit (Hint, Microchip do some really cheap low side dual mosfet drivers that can be used as small totem pole drivers for this sort of thing, quite useful), but here is one idea.
From Here:https://robs-blog.net/2018/03/31/radio-fundamentals-tx-rx-switching/
I built this and it worked rather nicely even with crappy parts. You actually want the resistors and bypass caps as well.
Thanks all for the replies. I took them onboard and decided to knock up a test board for this. Firstly some thoughts:
1. dmills solution solves a number of the problems but it does unfortunately increase cost and complexity. I need to replicate this circuit 10 times on one board.
2. Point about the elecraft TX switching using 1n4007 is interesting. I will look into that in detail.
Anyway, I built a test board and tried it with one power level and two sets of diodes to start with just to get a finger in the air. I will try it with different power levels. I am devoid of an SA so I'm not sure, past a cursory "that still looks like a sine wave", if it's adding any harmonics. I'm using a W7ZOI (AD8307) based log power meter here and Rigol DG1022Z AWG as the signal source.
diode=fairchild_1n4148, freq=7MHz, power_in=0dBm, insertion_loss_on_12v=3dB, insertion_loss_off_0v=60dB
diode=tayda_1n4004, freq=7MHz, power_in=0dBm, insertion_loss_on_12v=0.3dB, insertion_loss_off_0v=29dB
Interestingly the 1n4148 does pretty well if you can eat the 3dB of insertion loss. Now I'm going to have to do that twice and the filter has a 2dB insertion loss as well so 8dB down. Not a big issue as I can get the source to -5dBm and then stick 20dB amp on other end of it
TODO:
1. Test some non hooky 1N4007s (ordering now)
2. Test some 1N5711s (also ordering now)
3. Try different power levels (-10dBm)
4. Try different frequency (3.5, 10, 14, 21MHz)
5. Try optimising current for different diode types.
6. Try reverse biasing the bias supply (need to dig another power supply out of cupboard for that)
Edit: just looked at the KPA100 manual from Elecraft. That uses 1n4007 as TR switch as mentioned. Needs 90-150V bias which is pretty high. Interesting design!
I'd consider the 'standard' DPDT PCB relays, they're not expensive and it's only 12MHz so they should be OK, but it seems to me you should only need to reverse bias the diodes to get better isolation and you can do that by grounding the anodes and feeding a +12V supply to the cathodes of the diodes...
Adding C3 and C4 drops leakage further.
I actually worked out I need a metric ton of relays for this build though. It's a 5 band SSB/CW rig (remember that "simple SSB" thing we were talking about - it got bigger
). As it's got premix from the VFO and 5 TX filters that worked out at 25 relays in the end for the whole thing. And also on top of that they either have to be energised which brings default state problems and reliability into question or latching relays which are expensive.
Will have a play with your design as well - looks interesting.
Is a reason you do not use solid state rf switches, from Analog Devices or Peregrine?
I'd consider the 'standard' DPDT PCB relays, they're not expensive and it's only 12MHz so they should be OK, but it seems to me you should only need to reverse bias the diodes to get better isolation and you can do that by grounding the anodes and feeding a +12V supply to the cathodes of the diodes...
Adding C3 and C4 drops leakage further.
I actually worked out I need a metric ton of relays for this build though. It's a 5 band SSB/CW rig (remember that "simple SSB" thing we were talking about - it got bigger ). As it's got premix from the VFO and 5 TX filters that worked out at 25 relays in the end for the whole thing. And also on top of that they either have to be energised which brings default state problems and reliability into question or latching relays which are expensive.
Will have a play with your design as well - looks interesting.
25 relays is a lot, have you got separate BPF/LPF RX and TX filters?
Alan Wolke W2AEW has a nice video with various diode switching schemes, there's a way I'm sure, to shunt the signal to ground with diodes too but my head is filled with debugging documentation for a rollout project and a few other bits right now so I can't quite bring it to the front, I have a feeling it's in Alan's video, this one links to another:
Yeah it is a bit of a monster this one. It's got an analogue VFO which is premixed with a crystal oscillator and then bandpass (IF=4.915 MHz, VFO=2-3MHz, PMO=switchable), plus 5 RX filters, 5 TX filters. RX filters are double tuned. TX filters are low pass. I'm building it modular as I don't have time or headspace to think about anything other than small modules at a time.
Will watch W2AEW's video. I didn't actually notice that one. I've watched most of his over the last couple of years.
Quick update on this. Turns out that when switched on, the response of the test rig I did is flat to about 80MHz which is pretty good considering the construction. However when off, things are a little bit of a bastard. Turns out the inductors I'm using which are cheap axial lead ones are self-resonant at around 3.6MHz. I assume this is shifted a bit further up where things go awry to around 4.5MHz with a few pF of stray capacitance in the fixture and the diodes. Further investigation afoot at weekend as I've got to shelve this now.
Test rig!
Suck
... "ringing" the inductor with a square wave through a 100K resistor in series...
This stuff is great fun however. Shame about the day job which isn't!
A couple things:
- 1N4148 isn't a PIN diode, and you'll need quite a slow diode to do at 7MHz. The difference is minor at small signal levels: as long as you supply enough bias so that diode current doesn't drop significantly, it will maintain a low dynamic resistance. If you encounter more dynamic range however, or need to switch more than low dBm's, PIN diodes are desirable.
- The on-off ratio at that frequency and system impedance, should be alright. 1N4148 is a few pF off, and a few ohms on. Consider these quantities in your design.
- If you need more isolation, consider a tee switch. Invert the control signal and apply this to a diode in the middle, to short out the leakage.
- "RF" relays are expensive because they're coaxial and good for as much as the transmission line they fit to. You don't need that for mere HF and VHF. Regular tiny general purpose/signal/audio/telco relays will do. The on-off ratio is fantastic (low pF || Tohms off, mohms + nHs on).
Tim
Thanks for input. Much appreciated. Will read it carefully when I get some time this evening.
I am attempting to design a diode switched bandpass filter as relays are expensive and unreliable. I figured I'd test the switch first in LTspice. This is HF only (max 12MHz) so nothing too complicated. I'm currently here:
Input power is -7dBm from the LO premixer so I don't push the diodes too hard. There will be a broadband post-filter amp to knock this up to 7dBm to go into first IF mixer.
I tried biasing the V1 source with the following voltages and get the following associated loss:
12v (on): -2dBm - fine
0v (off): -35dBm - not great
-12v (off): -50dBm - good
Does this sound reasonable? Are there any better ways with better leakage / isolation?
I am mainly concerned about avoiding a negative rail to get better isolation if possible.
Edit: to note I changed the diode bias to 10mA after doing the above. Better linearity.
1. In your circuit, the isolation is degraded at 0V (compared to -12V) because the diodes rectify the RF and the resultant DC partially turn on the diodes.
2. To eliminate the requirement for negative bias, change the diodes to a long lifetime (tau) type. For your 3.5 MHz minimum freq, the diode should have a tau of >= 455 nS. The long lifetime diodes will be unable to rectify the RF and so will not develop any bias to turn themselves on partially.
3. Since you planned to order some diodes, why order conventional PN diodes? I second T3sl4co1l's recommendation to get PIN diodes.
4. Why did you choose to connect the diodes in the anti-series configuration? Unless the two diodes are matched, they will draw different current. This problem can be obviated by connecting the two diodes in series with same orientation; i.e. one current flowing thru both diodes.
5. However, the best isolation (wideband) can be obtained by alternating series and shunt diodes. When the series diode is OFF, the shunt one is ON. (see attached image)
Thanks for suggestions.
Not too concerned about 100% perfect isolation here. 60dB is enough and that's what I'm observing with 1n4148 diodes. Turns out negative bias doesn't make a whole load of difference to these. About 3-4dB.
While PIN diodes may be more suitable and have better overall characteristcs, I don't want to knock the price of this up too high as it's a one off. With respect to the normal diodes, they are general purpose so wil lbe used elsewhere anyway. I'm interested to compare insertion loss and total isolation on heavily doped (1n4007) with high transit time versus schottky vs signal diodes. This switching approach scales down to audio as well which is something I need to consider later on.