Also I need to attenuate from 0dBm down to about -100dBm i.e. down to almost nothing. It's obvious why those chips only go down to 30db attenuation: the capacitive coupling across the 5x5mm package is going to set a lower limit!
To do this with the above chips one would need four of them.
I would think that using a resistive ladder with MOSFETs ought to be the simplest way, especially as it can be stretched over a few inches of a PCB. Are FETs no good, due to too much capacitance? I am thinking of say a BSS138.
Signal generators use a string of PIN diodes but the resulting circuit is absolutely horrid... but they need to go to a few GHz.
If you go to an amateur radio supply store, you'll find that they don't sell attenuators that are good for more than 60dB. That's because trying to shield to more than that is a royal pain in the butt.
Chances are that unless your cables and terminations are perfect, you'll get more than 100dB of signal out of an empty box with shielded terminations on the plugs.
--
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com
$12 is expensive? Man, you are more of a penny-pincher than I am. Hobnestly, I don't think you can build this for less money. At least not in quantities under a few thousand.
I've got HP-355D attenuators here. They are selectable from 0dB to 120dB in 10dB steps, rated DC-1GHz and 500mW CW. Sometimes I cascade them with their 0-12dB brethren if I need fine steps.
That wasn't really my question. I was asking why the "obvious" method is not being used. The FETs cost about $0.05 each and an attenuator ladder could be made for under $1 in parts. There is presumably a reason why this is not being used.
Because the capacitances of the FETs mess with the flatness of the frequency response and their resistance upsets the attenuation value and impedance match. If you can live with that, fine, but for many applications that's just not good enough.
Serious wideband step attenuators are invariably mechanically switched T or PI resistor stages. The bigger steps are often built like two or more T or PI stages in series. You can't just use any old resistor that happens to have the right value, either. Resistors for use at RF are *made* for that purpose, lest parasitic effects take over at higher frequencies. Even so, attenuators are often full of gimmicks to flatten the response still further. That's what makes them expensive.
The last time I bought attenuators, I paid $250/pc, and thought that quite normal. Those were *fixed* attenuators. They are specified to within 0.5dB up to 18GHz and that has a cost.
It probably is being used, on that chip. You can get much better parasitics on a chip so if anyone wanted to use this circuit they would use something like that chip. If they had a big enough market and didn't like the price of that chip then they could do their own chip and the unit cost could be lower. I have used tapped resistive attenuators on cellphone chips, with MOS switches.
It is nice to be able to choose the size of the FETs, as there is a trade-off between on-resistance and off-capacitance, and so for a given resistive attenuator impedance, there will be an optimum MOSFET size. Newer chip processes give a better trade-off, though the allowable voltages then get smaller. For your suggested frequency range to 500MHz, I would suggest using a process with a gate length of 0.35um or smaller.
If you wanted to try making one of these circuits on a PCB, then I would suggest using NMOS bus-switch chips like the FST3125 or similar, (which radio amateurs sometimes use for switching in filters, and for building mixers). At least then you are free to bias the substrate a volt or so below the ground of your resistive attenuators, which would make the substrate diode less annoying than it is for discrete MOSFETs. You might need to make a T structure out of three FETS if you want good off-isolation. The two devices in series at the top of the T are turned on when the switch is passing signal, and they are both off and the shunt device (to ground) is on when you want to block the signal. You can put several output arms on a single T-switch if you want to get fancy.
If you don't need the minimum attenuation to be close to zero dB, and either the input or output doesn't have to be matched to 50 Ohms, then you can do other tricks to reduce the number of switches. For example you can make a R-2R ladder, and if you terminate it properly on both ends, (one end being the output), then you can feed a signal current into one tap along the ladder, and by switching to a different tap you get a nice 6.02dB step. If, instead of a R-2R ladder, you use a R-75R ladder, then you can have 1dB steps. You can extend the usable frequency range over which the steps are accurate, by observing which resistors have the parasitic capacitance of a FET in parallel with them, and putting tiny capacitors in parallel with the all of the other resistors so that the impedances in the attenuator are still in the correct ratio even when the frequency is so high that the parasitics are significant. The frequency response might not be flat up to a high frequency, but the steps can still be accurate up to a high frequency, and you can often correct the frequency response using some sort of look-up table if you know the signal frequency.
By feeding in an input current from a high-impedance source (such as a transisor collector), you can make the on-resistance of the switches relatively unimportant to the attenuation step size. You can also use the MOS devices as cascode transistors if you set up the DC bias differently.
I did wonder how cellphones control their power output. They obviously don't use an attenuator because that would hammer the battery - to run all the time at max power.
I looked it up. A very handy device. By "substrate" you mean the GND pin?
I need to visualise that :)
That will be OK because I will have plenty of spare power.
My plan was to production test the output at every step and load that into an EEPROM table, and correct for it thus.
The Peregrine chips are not that expensive but I would need at least 2 of them which does get a bit expensive, and I *think* that their max attenuation of 30db (60db for two) is going to be a bit of an illusion
*because* their small size prevents the attenuator to be stretched out and have individual parts shielded from others. Whereas a discrete solution can be spread out as desired. I have some experience of shielding parts of circuits, using little cases produced by chemically machined brass.
You mean the output of the "power amp" (say +3dBm) would be a current source?
As Chris was saying, it is likely that the IC does this with FETs. But this is not as easy as you may be thinking. You can't use a single FET, you normally need two per switch because of the parasitic body diode. Per attenuator section you need three of those combos, two for the path and one for going around. Times the numer of stages. With the requisite precision resistors and miscellanea you'll very soon exceed the cost of the chip with your discrete solution.
Also, even at 300MHz the gate caopacitances do get in the way. On an IC you have much more options to combat that.
For that kind of range i would use relays and shielded compartments. The relays would do 20 dB steps, below that you can use a variety of technologies.
The attenuators, if any, are always before the power amplifier, to avoid wasting lots of RF power as you suggested. Sometimes they are operating at at a lower frequency, though I have used them at RF.
formatting link
[...]
The substrate is the thick wafer of silicon (250um or so) on which chips are made. The transistors are made of very thin layers on the surface of the substrate, and they are isolated from the substrate by diffusions that make diodes which in normal use should be kept reverse-biased. Therefore if the substrate is p-type you should connect it to the most negative voltage in your circuit, and for n-type substrates it would need to be connected to the most positive voltage in your circuit to keep the unwanted diodes reverse biased. In discrete MOSFETs I think the substrate is normally the drain, and this is the cause of the annoying unwanted diode from drain to source. There are exceptions to the above, such as the chips from Peregrine where the MOSFETs are often dielectrically isolated so there is no substrate diode, and less capacitance.
[...]
I think without any special precautions you would get about 40dB - 50dB isolation from pins on one side of a 5x5 QFN to the opposite side, up to a few GHz. You can simulate it with FastCap and FastHenry from MIT, if you are very patient. If you solder little metal walls onto the PCB and add lids to make shield cans, then I think you could achieve a lot of isolation between different shield cans. As long as you are not looking for too much isolation within a single shield can, you should be able to achieve good isolation overall. Be careful not to share a single power supply rail between the different attenuator stages. Use filtering, separate regulators, etc. as necessary. Similarly put plenty of filtering on the control lines and re-buffer any logic lines using the clean power supply for the buffers if necessary. All of these precautions are used on cellphone chips to prevent coupling from one circuit to another.
That is one option which might work for your situation, perhaps. You can feed such a signal current into a tap along a ladder attenuator, and if it is a good current source with high impedance then it won't change the impedance looking into the end of the attenuator ladder. The main advantage I see is that you need only one on/off switch per attenuator stage, instead of two change-over switches per stage as used in a typical mechanical attenuator. On the other hand the off-isolation of the switch matters a lot more so it might not be good in your application. I like that approach more for baseband circuits where switches that are off can have really good isolation.
A few of those Peregrine chips with little shield cans soldered onto your board would be my recommendation. At 300MHz you shouldn't have too much trouble. If you can find a chip with e.g. a single 0dB / 20dB step, then that might better for most of the stages, as you only need one finely adjustable chip. Having the capability for fine adjustment must add a lot of extra switch parasitics.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.