Mechanical Design for good RF Performance

Created; 09/02/2015, Changed; 19/02/2024, 06/02/2024

Previous page; Engineering Management

The sound reproduced is warm mellow, as if there were a large loudspeaker and substantial baffle (Valve sound, which I think, was developed by Philips in the 1930s).  The inherent triode valve distortion was used to create sub-harmonic distortion that gives the impression that the radio has a better low frequency performance.

The circuit is identical to a 240V AC/DC model radio, although the rating plate on this radio specifies AC mains electricity supply only.  A lot of these cheaper radios had a heater transformer, but the most basic types like this one just uses series resistors to power the valve heaters.  The main input reservoir/smoothing electrolytic capacitor could explode, because of its age and because it has not been used regularly.  Old capacitors do not have a safety pressure relief vent, unlike modern electrolytic capacitors.  These live chassis radios are not safe, so be careful not to touch any metal in contact with the chassis, such as a screw within the control knobs.

The Wireless Museum in Dulwich, West London is well worth a visit, I visited in the mid 1990s.  They had working self-seeking Radio, I had never seen one of those working before my visit.  Working 405 line TVs, World War One valve manufacturing.  The components are so fragile because they were cheap at the time, but the museum manages to maintain these Radios.

Electromagnetic Compatibility;

There are a number of strategies for minimising radio emissions and susceptibility to radio emissions.  I intend discussing these points using various examples and small detours into other areas of electronics.  It is necessary to separate circuit function, stability and electromagnetic comparability, though the symptoms can be similar.  These things can interact with each other.

• The Ultra radio has 0V bonded to the metal chassis everywhere.  It is therefore filtered and screened very well within itself, but then it is connected by a mains cord forming star point - no other way is possible or necessary.  This is important to bear in mind, and I discuss screening and filtering later.

Similarly to the radio with PCB design; Contain any susceptibility or emissions locally (to the PCB) by;

• Use an embedded microcontroller so that all high frequencies parts are contained within the IC, rather than use a bus memory and microprocessor solution. 

  • This also has the advantage in that, for the money you also, have a fully tested function block.

  • The highest frequency outside the IC may be the crystal oscillator of say <5MHz, rather than a bus of say >25MHz within the IC. 

  • Other high frequencies could come from output pins with some embedded microcontrollers and may require 50R series resistors in the output to mitigate this.

  • If ball grid array IC are used then provide an 0V screen grid between the BGA pins, I have not designed anything using BGA and I do not know how well this measure mitigates given that the tracks are necessarily very thinner?  See http://www.cherryclough.com/ - **** I have not found an on-line copy of the EMC UK's journals in which the article was written ****

• Bandwidth limit the circuit using active and passive filtering.  This is a cheap option, that should be done anyway! 

• The PCB should have at least an 0V plane.

• Using PCB metal canning but do not just rely on the press fit contact but solder all around the lid (depending on the Can).

• Screen enclosed switch-mode power supplies and other power circuits, depending on the environment and the standards you are working to.  This may not be vital.

  • A 5mm clearance to the can or metal casing from an inductor matters a lot for a SMP <50W.  The clearance of 5mm is particularly low grade metal casting.

  • Bond or screwing to the PCB at 10mm intervals maximum is recommended. 

• Use differential signals rather than single ended signals where practical.  Even if it is not RF. 

  • On PCB's run two wires together around the board. 

  • Terminate the ends with resistor at output stages and capacitors at input, for filtering. 

  • Particularly if you are unsure or inexperienced at PCB layout. 

  • I make a point, to get it right first time and then cut back later, rather than get it wrong and fix it later, the latter way is commonly done but is more difficult. 

Filter (bandwidth limiting)

At the boundaries, which are the input and output signals and connectors, use passive filtering, LC or RC, as opposed to active filtering, this is very important.  Also keep the bandwidth limited to what is required throughout the circuit.  Avoiding high Q filter circuits, as they can ring at some frequencies (resonate).

• Active Filter; is a filter that uses an operational amplifier, capacitors and resistors to simulate an inductance, capacitor, resistor circuit or a high Q tuned circuit, but some passive filtering using capacitors and resistors should be included.  Active filters are generally designed to work over a limited frequency range, can be digital and can behave badly outside that frequency range.  Therefore, active filters should have passive bandwidth limiting to keep the frequencies within the range that the circuit can accommodate to avoid transient intermodulation distortion for example; - A fast signal that passes though an operational amplifier stage without filtering but causing distortion because the op-amp could not respond to correct it and quickly enough.  Digital such as microprocessor based filters with successive approximation analogue to digital converters will suffer severely from analysing noise if not preceded by passive filtering, you can see the effect when using a digital oscilloscope.

Transistorised Hi-Fi often used to suffer with susceptibility to lights and other thing being switched on or off, producing radio interference and causing clicks to be heard on the Hi-Fi.  Police and Taxi radio which used VHF band would "break through" on transistor radios.  The problem was that a bipolar junction transistor could rectify radio interference and thereby act as a detector (diode).  A fix that Electronics technicians, in the 1970s, used with amplifiers, was to fit a 100nF ceramic capacitor between the Base and Emitter pins of the first transistor.  I have also seen 100pF capacitor fitted similarly built into a design for an Eagle amplifier.

Positioning of components is also important.

 Small signals --- Switch mode driver IC --- Switching power and Inductor --- Input and output capacitors --- Filter section --- Input/Output connections. 

• The mounting holes - More could be fitted to improve EMC (a mounting hole every 10mms is recommended) but often what should be ideal is not practical and may not seem to work.  To avoid instability in the control loop, thereby contributing to EMC issues, the two mounting holes on the left may need to be isolated from metal work.  But if such a compromise is made there is a risk that another EMC issue will arise - this is the problem with reverting to such a star point strategy but bonding every 10mm is still much less risky than starting with a star-point 0V strategy.

• Signal and Power switch --- Low current and switching drive are kept apart.

• Circulating power around the Inductor --- PCB tracks need to be short thick and straight.

• Filter and energy reservoir --- For loop stability and the main part of the containing and smoothing out internal switching pulses.

• Segregation of internal RFI and external RF ---  I/O noise does not cross. Internal noise does not cross.

Adding a metal case makes electromatic emissions worse --- this is usually resolved by adding more earth points and greater distances between the inductors and the metal case. 

• Optimal ground point if multiple 0V connections don't work --- Is near the input and output.  What happens in practice can depend on apertures in nearby metal work and the qualities of the metal used in the casing.

Spot the omission - the circuit can have PWM driven to control the LED intensity without changing the LED's colour.  The control both isolates the LED using a transistor and also turns the regulator off during the off period.  The usual mechanical dilemma applies and there is no space to do what is desired and there is little or no overvoltage protection.  The solution is you have got to talk to the mechanical designers and stylists (the later people can stitch you up) both press your point and offer work-around's and discuss the problems with those or express the compromises. 

This sort of function can be placed on a PCB with other RF and signal circuits and although the power planes should not be split I've seen it done and the engineers who did that were not happy, but it was expedient.  In this case, the input output and power would be connected on the left, but an isolation to the planes put around the other three sides.  But if you can put the power supply block on corner edge and away from sensitive circuit, that is better.  Put very sensitive circuit on cable and coupled through common mode choke then bonded to a metal case is very good.

• The PWM method of LED intensity control is used by the automotive industry to drive the red tail light / stop light function.  The EMC measures taken are also higher than commercial industry to meet the much automotive industry standards. 

• The power switch Q1 and Q2 are almost certainly not required Unless a higher pwm frequency (say 100Hz)  is desired. The regulator, evidently from the low value shaping capacitor, settles to the required current very quickly, so there is virtually no ramp up ramp down current during the PWM cycle.  This part of the circuit should therefore be removed.   Further to the point, a p-channel-JFET switch has not been included to isolate the compensation circuit during the PWM off period because there is no compensation network required with this current mode regulator.  

• The arrow interconnections are used to connect to other circuit sheets.  But I have used them in some cases to connect nets on this sheet.  They are not required, and I have not used them in all cases. 

An instrument I designed called a polarimeter required the motor encoder light source and display all be placed in one end of the instrument, but the photodetector amplifier board placed in the other end of the instrument.  This would have made the instrument misshaped.  The motor and encoder were placed near the detector, where place together but away from the display.  The instrument worked better than expected was very successful, but when improvement was required the design layout now prevented that from occurring.  The options remained were to enclose the amplifier within a mu-metal box, the metallised screening of the instrument was otherwise good.  The design idea was developed over a 20-year period, but when an instrument was envisaged that was developed over a period of two years.  In the meantime, other more manually operated electronic polarimeters were developed or improved.

EMC and compromise might be expedient - this works, but is not recommended.  This sort of function can be placed on a PCB with other RF and signal circuits and although the power planes should not be split I've seen it done and the engineers who did that were not happy, but it was expedient.  In this case, the input output and power would be connected on the left, but an isolation to the planes put around the other three sides.

If you can put the power supply block on corner edge and away from sensitive circuit, that is better.  Put a very sensitive circuit on ribbon cable to a PCB and whatever sensor you are using mounted on that PCB is very good. Connecting the ribbon through common mode choke then bonded to a metal case is very good.  But also a star point but no common mode choke is, once again, not recommended and is also very good.

If necessary, a capacitor may be placed between the 0V plane and the connection to chassis to provide DC isolation (ensure that this part meets any standard for the industry and application).  But using the mounting points as shown this could increase emissions a solution is to fit a zero ohm link that can be removed but firstly add more mounting points and work on other things before resorting to isolating any mounting points.  RF emissions in the Near field are not predictable, and a metal enclosure for a power supply can make them worse.  Firstly, though, ensure that the controller control loop is stable that is optimally compensated or a little over damped.  If PWM for LED brightness control, ensure that the control loop does not RAMP but switches between on and off (this also ensure that the LED colour is stable).

• 1. The input and output are close together so that the DC current path is short and the AC path is very short (on the right-hand side).

  2. The tracks that carry current and particularly switching currents are short, fat and straight (Left side next to the IC).  These track widths can be varied over their length, the EMC performance might be improved by so doing.  But the effect of varying the track width is likely to be marginal, and in practice it is not done at these modest frequencies.

  3. This circuit is used to drive LED's.  In boost mode, the input current flows almost continually and the circuit is designed to have minimal switching frequency ripple.  In Fly-Back mode the AC current is also arranged to flow almost continually although the DC current is switched.  This Diodes Ltd. design should achieve a better performance than a SEPIC design, although there may be an increased current ripple on the LED passed through from the input supply.

  4. The transistor on the right provides PWM drive of the LED but avoids current ramping that would change the colour of the LED. (The LED colour changes with current, so dimming is achieved by pulse width modulating the LED current)

  5. See Diodes Ltd. data sheet for; ZXLD1374 https://www.diodes.com/assets/Evaluation-Boards/ZXLD1374EV2-User-Guide-Issue-3.pdf The first diagram shows the AC path with the output capacitor connected from Vout+ to 0V but the DC path is Vout to Vin+.  The PCB above also has protection against reverse supply with an input diode, this diode also prevents supply dips from causing over current pulses in the output LED.

  6. Figure 2 of the User Guide The circuit and component numbers are different on my PCB above which has an additional inductor between D1 and C5 and there is a transistor in the final output for PWM drive of the LEDs and another filter network in order to meet Automotive standards for emissions.

  7. There are a lot of via's from 0V on the top and underneath.  This is to make the 0V plane UN-fragmented and to provide as much copper in the 0V as possible.

  8. Place associated components together.  Keep path lengths short.  The main point here is that you have to prioritise and choose where you can make compromise.  Therefore, spend time on the PCB layout to minimise the track lengths, bends, and use thicker track where there are fast voltage and current transitions.  Ensure that the 0V return current can follow the same path as the signal current by using unbroken ground planes (the current will then find its own way back and at high frequency stick close to the signal track path).  That way, you can minimise the amount of compromises.  Also, using differential pairs for signals like transmission lines is an excellent approach in addition.

  9. Some capacitors and made up of three components, two large value capacitors plus one small NPO type to maximise filtering.  This is probably unnecessary one large, and one small value should be fine, and maybe just one large value capacitor will also be fine.  The switching power diodes are Schottky types, these have better switching characteristic that does not look like ringing compared to other very fast types of diode.

  10. PCB design - make a ground plane and spend time minimising and fragmentation of that plane.  If you need to add tracks in a ground plane, make them short and bring them back to routing layer(s) as soon as possible.  Therefore, minimise or avoid make holes in the plane layer.  Actually, the best strategy is to have two inner power plane layers, and two outer routing layers.  Putting the planes on the outside, apparently can cause buckling when soldering in an oven, but also components make holes in the planes which should be avoided as this would be fragmentation of the plane.  A top track layer and a bottom 0V plane layer works fine.

  11. One tip, if you can probe your circuit with an oscilloscope with the ground connection to anywhere on the 0V plane and there is no difference in the waveform observed, then you have understood and got a good PCB layout.  You should also find that pulse waveforms don't ring (overshoot many times), but some overshoot may be acceptable.  Think about this before you manufacture your PCB.  I shall discuss this later.

  12. Radio's like the Ultra above are physically laid out so that the RF input is followed by the mixer, IF amplifier, detector, Audio amplifier, to the output valve and finally the power input rectification smoothing and filtering.  So the power supply is near where the signal is largest at the output, and furtherest away from the RF input where the signal is smallest and most susceptible to interference.

Grounding strategy - Using one point (star ground) or grounding everywhere.

A star point, connection to a single point on a chassis, tends to give very good low noise performance, but there will be resonant frequencies and this is very difficult to deal with.  Alternatively grounding everywhere can give a higher noise level, because there may be varying current across the chassis that inject low level voltage into the circuit, but resonances are likely to be a lot less this configuration is currently more in favour.  I recommend that you pick and choose, but generally go with bonding everywhere.

For example, Radio's like the Ultra above, have a steel chassis which is signal ground (0V).  When ever 0V (possibly connected to mains live or neutral) is required, there is a short connection to a nearby point to chassis.  My homemade oscilloscope is also bonded everywhere (approximately - I was inexperience).  Therefore, neither design use much star point for grounding and both work well.  A strategy of using differential signals when bonding everywhere, like you would have to with Star point anyway, also improves the outcome.  You can place common mode chokes between PCBs and provided any common mode current due to voltage differentials across the metal work that would flow within the common mode chokes is less than the rated saturated current, there will be benefit.

Pictured below is an RF PCB layout for a very low power crystal oscillator and Phase Locked Loop of a microcontroller.  The PCB layout is an improvement on another Motorola (Freescale) example that used star point and made a point of having no 0V plane under the crystal oscillator section of the board.  But it still is not an optimal adherence to the rule of RF design should be bonded to 0V everywhere though it no doubt works, but I don't recommend it.  This design has completed the 0V guard track to form a ring around the crystal but only has one connection to the 0V plane - I don't see any merit in laying out the oscillator this way, but I have used an earlier similar reference design successfully.  Do ensure that there is an 0V plane as the next layer below the routing plane shown and that it is not fragmented, and don't place any other tracks under the crystal to prevent 0V return currents flowing under the crystal - the main point made in most applications notes. 

https://www.nxp.com/docs/en/application-note/AN2727.pdf (Page 21)

This application also suggests star point for the analogue systems, but it works fine with the 0V supplies taken by the shortest path to 0V plane and the supply decoupled and connected to the supply pin via a small series resistor, instead.  Not only is this approach I suggest correct, but it works.  On some occasions, though, you do have to follow the application note - particularly when there is a power supply start up order and the RC network on the analogue supply makes the order of start up wrong or that it simply won't work as well any other way.  The latter case comes about because (my theory) star point produces very low noise frequency bands and tuned notches as I say elsewhere, provided a tuned notch is avoided this may be the only way to achieve the performance required of some Delta Sigma or Sigma Delta A/D converter types.

A tip given to me for when it is necessary to use star point in an RF environment and that is adding a bead choke - this is a low Q constructed choke using a soft magnetic material such as used in mains filters.  Alternatively, add series resistors and keep the signal paths short and like a transmission line grouped together, following the same path.  Also, common mode chokes intended for filtering are similarly low Q and can be placed at the output or the input of board to board ribbon or multi wire cables and in that way each board can be connected to the chassis.

Examples of enclosures and magnetics;

Grounding everywhere (connecting 0V to metal work and bonding everywhere, not specifically earthing) is best, but grounding at one point has been common practice since 1960 until the turn of the century.  Power Inductor can induce current into the enclosure, so keep them some distance from metal enclosures.  It will be found that grounding at many points does not work in small enclosures and the workaround was to ground at one point where the power signals came in - So I use some star point if really necessary but not as a starting point. 

Keep some distance between wound components to minimise the coupling between them.  Some power input soft common mode chokes are wound with the former in a horizontal plane, whereas power chokes are wound with the former in a vertical plane in order to somewhat null and thereby minimise the coupling between them.  It may be possible to have three wound components in the three possible planes, but it is better to put distance between those components you can use both approaches.  The PCB layout for the LED PSU above shows wound components that are wound in the same plain, except for input bead chokes.  In that design it is not possible to source ideal wound components, but a design like this still can achieve very low radio emissions.  In the PCB design, I have phased the input and output chokes to behave as a common mode choke if there is coupling between them, but I have not experimented to see if there is a measurable benefit in doing that?

Useful Links;

• http://www.emcuk.co.uk/  Electro magnetic compatibility club.

• http://www.cherryclough.com/  I came across this company at Electronics Weekly sponsored show in June 2009.  They have an excellent understanding of EMC issues and solutions.

• My Blog for discussion - Redesigning a switch mode power supply based on a working example; Electronics - high frequency metal vapour arc lamp power supply

• Doulos training has some good sponsored webinars, I have this link from one webinar on high speed PCB design; https://speedingedge.com/

To discuss these electronics pages, see; Blog page Electronics 

Next page; Electronics - Electronics