High frequency passive low pass filter with cheap components

[suspension]

Hi all

I have a requirement to build a passive low pass filter with passband of 15MHz and a stop band attenuation of -70DB at-least at 30MHz. Can someone who has experience in this area please advise me if it is possible to build such a filter with cheap inductors/caps available from aliexpress which do not have any data sheet?

Thanks
Sus

[srb1954]

Hi all

I have a requirement to build a passive low pass filter with passband of 15MHz and a stop band attenuation of -70DB at-least. Can someone who has experience in this area please advise me if it is possible to build such a filter with cheap inductors/caps available from aliexpress which do not have any data sheet?

Thanks
Sus

What is the frequency that you require >70dB attenuation at?

The likelihood of success largely depends on how rapidly you want the filter to roll off to its stop-band attenuation. A filter that cuts off very rapidly will require high quality, precision components and you may not have much success obtaining such components from Aliexpress. However, a filter that has a much more relaxed roll-off rate will be more tolerant of lesser quality components.

[suspension]

I updated the question with that info. -70DB at 30MHz is what I am looking for.

[thinkfat]

Try this website: https://rf-tools.com/lc-filter/

Chebyshev topology, 6th or 7th order, choose "Standard" component values using E12 series.

PS: -70dB at 30MHz with a -3dB point at 15 MHz is very, very steep. It will take a lot of experimentation to achieve that with random Aliexpress components. I hope you have access to a VNA.

[BillyO]

70db per octave is a pretty tall order for a passive filter.  If you don't mind my asking what is the use case?

[suspension]

honestly, -70DB was bit arbitrary. All I want is to satisfactorily filter out 2nd+ harmonics from a signal which can have a max fundamental frequency of 15MHz. Apart from harmonics, I want to remove a 50MHz clock signal and its harmonics (which are clearly visible in scope and SA) as well.

I used a filter designer software and created an elliptical filter of 7th order and built it with aliexpress components. However, the response was nowhere close to a low pass. A forum discussion in a different site made me to believe that self-resonance is kicking in with inductors and caps making the filter useless. My only option seems to be to use branded components with low tolerances. However, many Chinese equipments seemingly built with cheap unbranded components have this sort of filters with satisfactory performance . I thought of posting this question here to get some idea from people who may be experienced in this type design process.

[mawyatt]

What you experienced is filter response element sensitivity, the Elliptical types are very sensitive to component variations/features. Chebyshev and Inverse Chebyshev are also sensitive, Butterworths & Bessels are less sensitive and can usually tolerate reasonable component variations, however have shallower transition to stopband.

Generally the steeper the passband to stopband transition the more sensitive to component effects. If you can tolerate a higher insertion loss then a pair of lower order LPFs separated by a attenuator can realize a good passband to stop band transition with standard components. You can't just cascade passive filters as they need to see a lossey source and load impedance to operate properly and why the attenuator between the two filter sections is necessary.   

Best,

[suspension]

wouldn't the sensitivity of component values result in a filter that has different stop/pass band frequencies but still has a shape of a low pass? In my case however, when plotted with a SA, I see a wired curve which looks arbitrary and has no resemblance to a low pass/hi/band pass filter. All I see is a response curve which reduces to around -20DB around 8-12MHz and increase to around -17dB above 30MHz and stays there untill 100MHz.

[mawyatt]

Depending on how far off the components are from "ideal" and what other artifacts they inhibit (SRF, ESR, ESL), and the type filter, you may see all kinds of strange behavior.

As mentioned the Elliptical types are very component sensitive, they utilize transmission zeros in the transfer function to create the stepper stop to passband transitions, which tend to be very component sensitive in frequency location in the stop band and thus have a strong effect on the pass to stopband transition, not to mention how "deep" the transmission zeros achieve.

You likley will achieve a better representation of your simulated ideal filter response with a Butterworth or Bessel type. We utilize the Butterworth filter types often, because it usually tolerates non-ideal components well and these often can be standard values without significantly altering the filter response.

Best,

[suspension]

Thanks, will try out a Butterworth or Bessel type and see.
Do you think self-resonance can come in to play at these frequencies/inductances/capacitances (2.2uH, 50-500pf) ? I guess if that's the case, even Butterworth or Bessel types wouldn't give desired results?

[T3sl4co1l]

Backlink for those curious, https://electronics.stackexchange.com/questions/647769/elliptic-filter-implementation-deviates-from-the-design-and-simulation
At least, it sounds like the same question.

A good idea is to find models for all components, as well as you can; and tweak the SPICE model until it matches your observations.  The problem with inductors of that type is they probably have a low SRF so you're dealing with not just equivalent R and C but higher order (series resonant or other TL like effects) modes too.  And power inductors, if they have models or characteristics at all, likely don't show those higher modes at all.  (For example, Coilcraft parts typically end just before or just after the first (parallel mode) SRF.)

Like, I made this filter,



Left side is a VCO, the right 10k 1/4W axial resistor is R8 and so on to the output at the right.  So it's filtering the VCO output.

Equivalent circuit,



with design response,



or zoomed out,



Notice that copper is removed around L2 and L3, and gimmicks have been added (bent wires, bits of copper clad).  C4 is the varicap.

C6-C2-C5 is an unavoidable capacitor divider, necessary to get the lowpass edge but the stray capacitance (along with the inductors' capacitances) causes an impedance transformation.  In this case that's helpful, because Q1 output impedance is quite high, and it's just signal level so I can get a bit more level out that way.  (The impedance at VC1 is, err what was it, a kohm, or more like 10? I forget.)  The impedance at Q2 base is fairly low (50 or 500 ohms? I forget, again); it's also adjustable by R13, if you don't mind the change in output gain as well).

I don't have a proper spectrum of it but this was the amplitude while sweeping the VCO over its range.  So the axis is nonlinear, but you can see the peaks and relative flatness:



Hmm, that must've been before I got my spec... circa 2017.  Hmm no, I got that 2015, why didn't I plot it that way?... :shrug:

Anyway, as you might guess, this model is quite a bit different from what you'd have designed normally.  This is at 100MHz so of course RF type inductors are used (SMT chip air core) and their behavior is pretty simple, but at somewhat lower frequencies you'll still want to use similar types -- air core is still suitable (give or take Q, you might be looking at needing quite high Q here) or you might use ferrite or powder types, usually still in single layer windings, chip style. 

Tim

[srb1954]

Thanks, will try out a Butterworth or Bessel type and see.
Do you think self-resonance can come in to play at these frequencies/inductances/capacitances (2.2uH, 50-500pf) ? I guess if that's the case, even Butterworth or Bessel types wouldn't give desired results?

You are going to need a fairly complex 12th order Butterworth filter to meet your stop-band specs but meeting those same specs with a Bessel response is going to be practically impossible due to the very slow roll-off rate of a Bessel filter.

Self resonances will still affect the response but if you can chose your inductors such that their SRF is well above your desired stop-band then that effect should be minimised in your frequency range of interest. However, continuing to increase the frequency may result in the filter attenuation actually reducing due to SRF effects where your inductors stop behaving like inductors and start behaving more like capacitors and start passing the higher frequencies. This may be a problem if you still have significant frequency components well above your stop-band that you still need to get rid of.

[thinkfat]

I also suggest you think of avoiding magnetic coupling between the inductors of the filter. I've seen discussions here in the forum about that aspect, keep the individual filter stages well separated. For example, don't put the inductors all in a straight line, instead angle them 90 degrees against each other or apply other tricks to isolate the stages.

[T3sl4co1l]

Interestingly, that might help, since it does a similar thing to the capacitors.  It's not going to be a calibrated amount, though(!).

Tim

[suspension]

Just to understand this bit more, I built a simple air core (not surface mount) inductor (See below picture) and measured its inductance using the method described here: https://www.youtube.com/watch?v=iQQe8uSZ8xc. This gave me roughly 0.000887808 mH. Used this inductor in the following simple low pass filter and simulated. Simulation shows a fairly ok low pass response. In this circuit, R1 represents the internal 50 Ohm resister in function generator, R2 represent the input impedance of my TinySA spectrum analyzer. C1 is the external cap.

Buit this circuit and plot the response using TinySA and this also does not show any low pass profile. (Please see the photo).

I am not sure if this is also due to low SRF of the inductor I built?

Would the inductor with digikey part number LQG15HZ2N2S02D work for my purpose?

[807]

When you experiment with circuits in the range of RF frequencies that you are playing with, you need to minimise stray capacitance & stray radiation.

The input to the circuit needs to be fed with 50 Ohm coax, not unscreened leads. Same with the output lead to the SA. Reduce those unscreened leads to an absolute minimum, or even better, use a piece of coax soldered to the board with an SMA plug on the other end. Get rid of that big piece of copper & solder all earth connections to 1 point.

[MrAl]

Hello there,

The actual success of this project as stated seems very unlikely.  Perhaps with relaxed specifications.

A 70db cut from a frequency F to a frequency 2*F (as you stated 15MHz to 30MHz) means a 70db per octave roll off which is probably impossible.
Maybe using two or more filters with buffers in between might work.

You have to realize that a first order filter gets you -6db per octave and Nth order filter gets you N*6db down, so considering 1st is -3db and 70db would require 67db that would take you to around an 11th order filter to get -70db down at one octave.

The clock frequency is a little easier to get rid of but not much.  It's roughly 4 times the fundamental which is better than 2 times.
Using a 2nd order filter gets you down to about -24db and -20db is usually good enough.

You have to realize that -70db is a cut by a factor of about 1/3000, while -20db is a cut of about 1/100.  Big difference.

Some of the DSP type sine wave generators use passive filtering on the output to get a cleaner sine wave from a PWM digitally generated signal.  You could check into those to see how they manage the filter design which is higher order.  They post the schematic online but it is interesting to see how they did the component layout also on the PC board.  If you cant find one online i'll look up a part number for the board and post it here.

[T3sl4co1l]

Note that your circuit looks more like this, with ballpark values added, and some coupling between the pairs of inductors (say k=0.1 for the two on the right, k=0.3 on the left??).  Further, they will be good antennas, and pick up ambient fields, of which, you're probably seeing commercial FM or TV in the 100MHz range?  That will mostly be picked up by the longer wires, then the shorter ground wire acts as an impedance divider to it, of course still bringing plenty of signal into the receiver.

The ground node is meaningless, just a reference node for SPICE purposes of course; the voltage on that PCB with respect to the surroundings may involve common mode signals from both the source, and induced from ambient fields.

Ground becomes more meaningful if you solder that coax connector right onto the board, all ground pins touching it; or better yet, on its side so a complete face is joined all around.  That essentially eliminates the ground side inductor, and the signal can be routed close to the board, greatly reducing coupling from ambient fields, greatly increasing CMRR, and giving its length a more controlled impedance (round wire over ground plane, at a height equal to typical hookup wire insulation, gives something like 60 ohms characteristic impedance).

Tim

[suspension]

Thanks everyone for all the valuable input.
I still do not understand why none of these problems seem to arise for the circuit explained in the following video.



He gets a nice looking low pass response with -3dB at 7MHz. He does not even know the value of his inductors and from which material the core is constructed.

[suspension]

Hello there,

The actual success of this project as stated seems very unlikely.  Perhaps with relaxed specifications.

A 70db cut from a frequency F to a frequency 2*F (as you stated 15MHz to 30MHz) means a 70db per octave roll off which is probably impossible.
Maybe using two or more filters with buffers in between might work.

You have to realize that a first order filter gets you -6db per octave and Nth order filter gets you N*6db down, so considering 1st is -3db and 70db would require 67db that would take you to around an 11th order filter to get -70db down at one octave.

The clock frequency is a little easier to get rid of but not much.  It's roughly 4 times the fundamental which is better than 2 times.
Using a 2nd order filter gets you down to about -24db and -20db is usually good enough.

You have to realize that -70db is a cut by a factor of about 1/3000, while -20db is a cut of about 1/100.  Big difference.

Some of the DSP type sine wave generators use passive filtering on the output to get a cleaner sine wave from a PWM digitally generated signal.  You could check into those to see how they manage the filter design which is higher order.  They post the schematic online but it is interesting to see how they did the component layout also on the PC board.  If you cant find one online i'll look up a part number for the board and post it here.

Thanks for pointing out this. Yes this may be too steep. But what I am trying out now is a pretty simple 2nd order low pass filter with one L and one C just to see a low pass response without much regard to the actual slope. Still what I get is garbage - mostly a random response.

My actual use case is actually to Filter out various components from a DDS generated sine wave. This includes quantization noise, phase truncation noise, clock signal and its harmonics, non-linearities in DAC. If you can post a schematic of an output stage of such a system, that would be great.

[suspension]

Note that your circuit looks more like this, with ballpark values added, and some coupling between the pairs of inductors (say k=0.1 for the two on the right, k=0.3 on the left??).  Further, they will be good antennas, and pick up ambient fields, of which, you're probably seeing commercial FM or TV in the 100MHz range?  That will mostly be picked up by the longer wires, then the shorter ground wire acts as an impedance divider to it, of course still bringing plenty of signal into the receiver.

The ground node is meaningless, just a reference node for SPICE purposes of course; the voltage on that PCB with respect to the surroundings may involve common mode signals from both the source, and induced from ambient fields.

Ground becomes more meaningful if you solder that coax connector right onto the board, all ground pins touching it; or better yet, on its side so a complete face is joined all around.  That essentially eliminates the ground side inductor, and the signal can be routed close to the board, greatly reducing coupling from ambient fields, greatly increasing CMRR, and giving its length a more controlled impedance (round wire over ground plane, at a height equal to typical hookup wire insulation, gives something like 60 ohms characteristic impedance).

Tim

I've actually simulated this with LTspice and still get a nice-looking low pass response (please see attached). But in the actual physical circuit, the response is random pass. That is what I do not understand

[joeqsmith]

...
But what I am trying out now is a pretty simple 2nd order low pass filter with one L and one C just to see a low pass response without much regard to the actual slope. Still what I get is garbage - mostly a random response.
...

Looking at your construction practices, I assume this RF stuff is new to you.   You may want to have a look at something like the ARRL handbook where they show various construction practices. 

Here is a simple LC filter you asked about.  I just grabbed an inductor and cap out of the drawer and stuck them onto a breadboard.  There was no attempt to calculate or measure the components and the VNA isn't calibrated.  The only goal is to show you a simple LCish network acting as a lowpass.

I use the term "ish" because while it looks simple, there's a lot going on here.   To give you some clue as to the effects of construction, in the attached video I am using a low cost VNA to attempt to look a simple RLC network.   It's built on a this same breadboard.   At 33:30, I show the network being measured.  I then cut the leads and remeasure.  Note the difference in results.

[T3sl4co1l]

Thanks everyone for all the valuable input.
I still do not understand why none of these problems seem to arise for the circuit explained in the following video.

He literally turns off the lights and the noise floor drops 20dB!  What a terrible setup, huh?!  (With solidly grounded coax connectors, and maybe some shielding over the filter, that would completely go away.)

I don't know why he doesn't know the material; yellow-gray or yellow-clear is a standard Micrometals part (and most other manufacturers seem to follow the same color coding): https://www.micrometals.com/products/materials/pc/ and https://www.micrometals.com/products/materials/rf/ I would guess #35 would give way too much inductance and loss, so it's probably the #6.  Which I think he said was in the kit?  So it would be weird if it wasn't?

And the valley at, what, 13MHz, then it rebounds!  That's clearly due to the capacitor lead length, and maybe something about the clip leads connecting to it.  It would be even worse if expanded to show the response at higher frequencies -- not necessarily more transmission, but more lumpy (uncontrolled) response.

The prototype also seems to be pretty gentle, Butterworth or so, plus the losses of the inductors (notice the 1dB or so insertion loss in the transition band i.e. 7MHz +/- a few times), which makes it more tolerant of errors.  The outermost capacitors are way out of tolerance, even!  A return loss plot would be more illustrative, probably showing where some of that insertion loss is coming from (i.e. it's reflecting off the filter rather than being passed through); best tuning is had by measuring return loss.

Mind, I'm just skipping through here, I don't feel a need to watch 28 minutes of something I've done many times before.  Let me know any highlights to look at.

Tim

[joeqsmith]

Outside of maybe learning some of the basics, you don't normally want to mix RF with breadboards.   Here's a low power 12MHz Chebyshev 5th order lowpass using three inductors and two capacitors.   It never had a purpose other than for demonstration and I never took the time to try and improve the ripple.  It has a nasty dip.   The filter is made using surface mount parts and is roughly 15mm by 10mm. 

[G0HZU]

Hi all

I have a requirement to build a passive low pass filter with passband of 15MHz and a stop band attenuation of -70DB at-least at 30MHz. Can someone who has experience in this area please advise me if it is possible to build such a filter with cheap inductors/caps available from aliexpress which do not have any data sheet?

Thanks
Sus

I think you have to define what you mean by cheap. If basic 0805 C0G SMD caps are allowed and Micrometals powdered iron toroids are allowed then 70dB rejection is going to be really easy to achieve at 30MHz whilst maintaining a low insertion loss. 90dB rejection at 30MHz would require an 11th order Cheby filter. You could expect to see about 0.7dB insertion loss at 15MHz according to a filter simulation based on 0805 C0G caps and toroid inductors. This is for an 11th order cheby filter.

The more challenging part would be maintaining a 70dB stopband performance out to (say) 1GHz. This would require a decent layout and probably a tiny roofing LPF if you were aiming/hoping for >70dB stopband performance out past 1GHz.