Essential scrap

Improved frequency reference for ELV FC-500

Wed, 29 Nov 2023 00:00:00 +0000

Improved frequency reference for ELV FC-500

As one of my first electronics projects, two decades ago I built a ELV FC-500 frequency counter out of a kit. It has worked great and found use in several projects where I have needed to check the accuracy of oscillators. I just wished for a bit more accuracy on the meter.

Original design

The FC-500 is based on a 24-bit ripple carry counter implemented with 74HC393 chips. A microcontroller controls clock gating and counter reset, and reads out the final count for display. The display has 8 digit resolution, providing display resolution of 0.1 ppm. This far surpasses my DS1054 oscilloscope, which only provides three digits in its frequency measurement.

Amazingly, the original schematics and build instructions are still available from ELV journal, split into part 1 (schematics) and part 2 (build instructions) . For the price of 1 EUR, I decided to buy both parts as PDF instead of trying to find the original magazines from my archives.

What I really like about this device is that it correctly implements input-synchronous sampling of counter. This means that it is able to display accurate frequency and period down to 0.01 Hz and possibly lower. This was very handy for calibrating a clock that gave pulse once a minute.

Improving the accuracy

The original design uses a basic 4.096 MHz clock crystal, combined with a trimmer capacitor. The capacitor allows adjusting the meter using a known reference source. However there is no temperature compensation for the crystal, which has about 0.2 ppm/°C slope. My FC-500 had a few ppm of inaccuracy in room temperature before I started this project.

For highest frequency stability, common choises are ovenized crystal oscillators and rubidium references. For a portable device like FC-500, these are too large, too power hungry and require continuous power.

Temperature compensated crystal oscillators are a reasonable alternative. For a device like this, you want a VCTCXO: VC for voltage controlled, TC for temperature compensated, XO for crystal oscillator. Voltage control allows adjusting the oscillator for initial calibration. Temperature compensation should keep it accurate after that.

What remains is the aging of the crystal. Typical VCTCXOs have ± 1 ppm per year aging specification. There are MEMS based oscillators that are much more stable than this, but also more expensive.

A complication is that VCTCXOs with 4.096 MHz frequency are quite rare. Fortunately 16.384 MHz is better available, and a divide-by-four circuit can be built with two flipflops. As possible options, I identified the following parts:

LFTVXO009905BULK : 14 EUR, crystal, ±0.9 ppm stability, ±1 ppm/year typical aging.
AST3TQ-V-16.384MHZ-28 : 18 EUR, crystal, ±0.3 ppm stability, ±1 ppm/year max aging.
SIT5356AI-FN-30VT-16.384000 : 73 EUR, MEMS, ±0.2 ppm stability, ±0.06 ppm/year typical aging.

As tempting as the MEMS accuracy is, I decided to go with the cheapest one for now. They are all pin compatible, so an upgrade is easy to do if I ever want to.

Divider circuit

FC-500 operates with a 5 V logic supply, so I needed a voltage regulator to provide 3.3 V for the VCTCXO. Other than that, there are the two flipflops for dividing the output frequency and a trimmer resistor for adjusting the control voltage.

The PCB is a 1-layer design, suitable for DIY manufacturing by etching or milling.

Installation

Many microcontrollers have two pins that the crystal resonator connects between: Xin and Xout. These are the signals of an internal inverter gate used for implementing a crystal oscillator.

When replacing a crystal resonator with an external crystal oscillator, the signal must be fed into Xin. This is the pin with the variable trimmer capacitor, if the circuit has one. You can also measure the amplitude with oscilloscope (with probe in 10x mode), where the Xout pin will show higher peak-to-peak voltage. In this case I was able to check the pinout from the original schematic.

I glued my small auxiliary PCB near the battery compartment, and ran a wire to the microcontroller that is below the LCD display. The original crystal has to be desoldered, but initially I left the variable capacitor in place.

This however caused a bit of a problem. The large capacitance at the end of a long wire, combined with high slew rate of 74LVC80 flipflops caused large spikes. Debugging this was complicated by the weird behavior of the MIC5504 regulator I originally had in the circuit. Apparently high frequency spikes can couple into its feedback circuit, causing output voltage to drop. They do mention in datasheet that "The MIC5501/2/3/4 is not suitable for RF transmitter systems.".

Adding a 100 ohm series resistor on the output and cutting the trace that led to the trimmer capacitor made things a lot better. I had already switched the regulator for XC6210, which proved stable even under the less than ideal conditions. A final issue was that I had to cut a small groove in the LCD display holder to have space to pass the wire underneath.

Calibration

I used a GPS module with pulse-per-second output as the calibration source. Careful adjustment got me to the correct 1 Hz reading, but there seems to be a small disrepancy in the last digit between period and frequency measurement modes. Probably this is a limitation of how accurately the firmware does mathematics. I guess the high-accuracy MEMS oscillator would have been somewhat wasted in FC-500.

Design files

The KiCad design files for the clock divider are available here .

– Petteri Aimonen on 29.11.2023

Comment on this page

TEM cell construction

Tue, 31 Oct 2023 00:00:00 +0000

TEM cell construction

Building a TEM cell is easy in principle: it is just a metal box. But it does require careful dimensioning to work well, and many of the formulas online are just plain wrong.

Impedance matching

The characteristic impedance of a TEM cell is the ratio of voltage and current at each point along its length. Things are simplest if it matches your spectrum analyzer, which typically have 50 ohm inputs. If needed, a transformer or a resistor circuit can adapt between different impedances. Mismatched impedances lead to reflections at connection points, causing attenuation or amplification depending on frequency.

The impedance depends primarily on the ratio between the width of the middle conductor (septum) and the distance to ground planes (shield). Wider septum and smaller distance give lower impedance. The thickness of the material and the width of the shield affect it to a lesser extent, as does anything placed inside the TEM cell. It is usual to aim for an impedance slightly higher than 50 ohms, as a device placed into the cell will reduce the effective distance to ground and thus the impedance.

The quality of matching can be characterized by VSWR measurement. Ideal match would have a VSWR of 1, while a VSWR of 2 means that reflections cause up to 2x difference in amplitude along the cable. In the context of EMC measurements, VSWR of 2 translates to 6 dB uncertainty in the measurement result. The amplitude error caused by impedance matching is frequency dependent, and while it can theoretically be calibrated away, it is easier to just try for a good match.

Even a perfectly matched cell will have resonant frequencies which will limit its usefulness at high frequencies. The first resonance peak will occur at a frequency where the width of the septum fits half a wavelength of signal. The TEM cell can be used even beyond this frequency, but the amplitude response is uneven, increasing the uncertainty of measurements.

Literature research

The history of TEM cells begins with the 1974 article Generation of Standard EM Fields Using TEM Transmission Cells by Myron L. Crawford. It details the basic structure of a TEM cell, its uses and the formula to calculate the impedance. Simplified for air as the material between plates and use of thin material, the impedance is calculated as:

Z₀ = 94 ohm / (w / b + C_f / 0.0885)

Where w is the width of the septum and b is the height of the cell. C_f is the fringe capacitance, which Crawford determined experimentally for a few geometries using scale models.

Crawford's results are excellent, achieving VSWR of under 1.2 up to the resonant frequency of the cell. In his case, the septum is 1.2 meters wide, leading to a low resonant frequency of 125 MHz. A smaller cell will have higher resonant frequency, but there is less space to fit devices inside. Fortunately the electronics we have in 2023 are much smaller than the electronics of 1974.

Skip forward to 2008. TEM cells are popular and widely used, but commercially available ones are expensive. Sandeep M. Satav published the paper Do-it-Yourself Fabrication of an Open TEM Cell for EMC Pre-compliance which used PCB material for the construction of an open TEM cell. Very good idea and easy to manufacture, and the construction method has found popularity in the DIY community.

Unfortunately there is a small mistake in the geometry calculations. Satav uses the rectangular geometry described in Crawford's paper, and the value C_f = 0.053 for fringe capacitance that Crawford determined experimentally. But Crawford used closed TEM cells, which have four metal walls. Satav's paper shows an open TEM cell with only two walls - easier to build and use, though less shielded from outside noise. The fringe capacitance 0.053 is an incorrect value for an open TEM cell.

The error in the geometry yields a characteristic impedance of about 70 ohms. While still usable, this shows up as less-than-ideal VSWR of over 2 in Satav's results.

Later in 2020, Michal Hrouda built a cell using Satav's dimensions, yielding similar performance.

Scale models

To verify that I was indeed correct about the mistake in dimensions, I decided to follow Crawford's example and built a scale model. For material I used cardboard and copper tape, with a printed template to cut them accurately to size.

For the first prototype, I chose the square cross section from Crawford's paper, scaled down to 40 mm shield width and 33 mm septum width. At first I built it without the side walls. With 50 ohm terminator at the other end, the S11 measurement done with NanoVNA V2 showed wildly varying impedance. This is because the TEM cell acts as a quarter-wave impedance transformer .

The 700 MHz peak frequency matches a quarter wave with the 100 mm length of the cell. From the 123 ohm peak impedance, we can deduce a characteristic impedance of sqrt(50 ohm * 123 ohm) = 78 ohm. From this, we can calculate the actual fringe capacitance value and find it to be 0.033.

To verify that open vs. closed TEM cell makes a difference for the geometry, I then added side walls and got characteristic impedance of 57 ohms.

Finally, I calculated new dimensions for an open cell with C_f = 0.033, built a scale model and got a characteristic impedance of 55 ohms. I adjusted my final estimate of C_f down to 0.03.

Size of the box

The larger the TEM cell, the bigger devices you can fit into it. The common advice is to occupy at most 1/3rd of the vertical space. And for proper coverage, you'd want to rotate the device into all three major orientations. For 100 mm plate separation, that would give maximum device size of just 30x30x30 mm.

I decided to disregard this, because I don't have the room to store a huge box. 100 mm plate separation would have to do. With the 0.03 value I obtained earlier, I calculated 300 mm as suitable width of the septum. I scaled the shields in same ratio to 350 mm wide.

Another size decision I made was to have the ends taper at an angle of 45 degrees. Shallower angle gives better impedance match in the tapered section, but it also increases the length a lot.

I had already bought some PCB material in three sheets of 610x457 mm. With the 45 degree taper angle, center section length would be 300 mm and each tapered section takes 150 mm of material, for a total of 600 mm. This allowed me to fit each of the three parts on its own sheet, minimizing cutting and joining of the pieces.

Cutting the sheets

For me, CNC router was the obvious choice for cutting the PCB sheets. But this could equally well be done with a jigsaw or just about any saw - a few millimeters of inaccuracy doesn't make much difference.

The material I used was 1-sided PCB sheet from TME . Solid sheet of copper, brass or aluminum would work equally well electrically. I have seen other projects use double-sided PCB and joining the sides by soldering wires through holes, but I don't think that is necessary for the power levels involved.

Construction

The shield pieces on top and bottom have a bend to them - 45 degrees in my design, 30 degrees in designs with a shallower taper. My plan was to route a slot almost all the way through the PCB material, to bend the sheet and use epoxy glue to fix the angle. This would avoid having to solder together multiple pieces and ensure a continuous copper surface.

Unfortunately the copper layer disagreed with me and ripped apart in multiple places. To fix this I used copper tape and soldered bridges every few centimeters. Another layer of copper tape on top, and it should bridge the pieces well enough.

I chose to use SMA connectors at each end. My initial plan was to just solder the septum to the middle pin and the shields to the body of the connector. I quickly found out that this put way too much load on the connector middle pin, and resorted to using epoxy to keep the septum in place.

In the center section the correct distance and alignment of the sheets is maintained by 3D printed side supports. I printed these at lowest infill setting to reduce their effect on capacitance.

In retrospect the best orientation would be to have insulating sides of each PCB towards the device being tested. This avoids any accidental contact to the conductive surfaces. I happened to solder the septum the opposite way around, but I can always add a plastic film for insulation.

Results

I used NanoVNA V2 to measure the cell parameters. I collected the previosuly published measurement results for comparison:

Generation of Standard EM Fields Using TEM Transmission Cells by Myron L. Crawford et al, 1974. Closed 4-wall 1.5 meter wide TEM cell made of 3 mm aluminum sheet.
Do-it-Yourself Fabrication of an Open TEM Cell for EMC Pre-compliance by Sandeep M. Satav et al, 2008. Open 30 cm wide TEM cell made of double-sided FR4 PCB material.
Tekbox TBTC2 , first released in 2016. Open 30 cm wide TEM cell, shields made of solid aluminum and septum from PCB material. Special construction of septum reduces high-frequency resonances.
Making TEM cell for EMC measurements by Michal Hrouda, 2020. Open 30 cm wide TEM cell made of 0.5 mm copper sheet.
Increasing the Test-Volume of Open TEM Cells by Using an Asymmetric Design , by Christian Spindelberger et al, 2022. Asymmetric open 30 cm wide TEM cell made of 3 mm aluminum sheet.

The results appear to confirm that the dimensions presented in 2. and used also in 4. are suboptimal. Tekbox TBTC2 is superior in high-frequency performance. In frequencies less than 500 MHz all correctly dimensioned cells give VSWR under 1.5, which is perfectly fine for all EMC pretesting purposes.

Further improvements

Christian Spindelberger's paper details an asymmetric TEM cell design, which has a higher lower section and thin upper section. This makes much better use of vertical space, as in a symmetric cell the upper partition doesn't have any practical use. I'm not entirely sure how the asymmetry would affect the sensitivity of measurements.

Tekbox uses slots in the septum copper, bridged by some kind of ferrite or resistive material. This dampens any perpendicular electromagnetic fields and reduces high frequency resonances. A similar technique is explained in "Expanding the Bandwidth of TEM Cells for EMC Measurements" by Myron L. Crawford, 1978 (DOI: 10.1109/TEMC.1978.303664), but instead of slotting the copper it uses absorbent material near the shield. Again I do not know if these methods will effect sensitivity of the measurements.

Design files

I have made a parametric FreeCAD design of the TEM cell I constructed. The main dimensions are in a spreadsheet element and can be adjusted to custom dimensions. The 3D model dimensions will then automatically update to match.

The design files are available GitHub and the basic dimensions also in PDF files linked below:

– Petteri Aimonen on 31.10.2023

Comment on this page

Integrated LISN

Tue, 31 Oct 2023 00:00:00 +0000

Integrated LISN

Measurement of conducted emissions is as simple as measuring either the voltage or current at the cables going to the device. Typically this is done for the power cables. Most common way is to measure voltage when power is supplied through a line impedance stabilization network.

The purpose of a LISN is to present a consistent 50 ohm impedance towards the device. Any noise currents then yield a predictable noise voltage that can be measured.

For very low power devices, a simple series resistor could be used. But most devices draw too much power to be supplied through a 50 ohm resistor, so an inductor is used instead. Common types are 5 µH and 50 µH LISNs, depending on the lowest frequency being measured. For pretesting, a 50 µH LISN can be used for all standards, as it covers the widest frequency range. Most LISNs also include a prefilter to eliminate any noise conducted from the DC power supply towards the device.

Traditionally LISN and TEM cells are entirely separate pieces of equipment. I chose to integrate a LISN directly to my TEM cell. This makes things simpler in several ways:

Supplying power through the LISN prevents external noise from affecting radiated emissions measurements.
Conducted emissions measurements are performed on top of a ground plate, and the TEM cell already has a ground plate.
Being able to quickly compare radiated and conducted emissions gives clues to where the noise originates from.

Design

I followed the basic design of LISN described in CISPR-16. A two wire LISN has two output ports. Depending on model they are connected either directly to the positive and negative terminal, or through a transformer to separate the common mode and differential mode noise . First type is simpler, but requires extra math on the measurement instrument. For easy use, I decided to include a transformer.

The input section consists of capacitors and a common mode choke. Their purpose is to block any noise coming from the power supply. Next are the series inductors. I used 2x 22µH on each line, for total of 44 µH. This is close enough to 50 µH for my purposes. A single inductor is fine too, but make sure to stay well below the saturation current rating of the components.

The output of the inductors goes straight to the connectors for the device being tested. The output is also connected through capacitors to the measurement instrument. In the simplest form, each capacitor would go to a coaxial connector, to which the 50 ohm impedance measurement instrument would be connected. At high frequencies, the series inductors have high impedance, while the capacitors act as shorts. For example, at 1 MHz a 50 µH inductor has 314 ohm impedance and 1 µF capacitor has 0.1 ohms. This arrangement directly presents the measurement instrument's 50 ohm impedance to the load.

But instead of measuring each wire separately, we want to know if they fluctuate together (common mode) or inversely (differential mode). A transformer can be used to calculate both the average and the difference between the channels.

Finally, there is a need to protect the measurement instrument from sudden transitions. For tinySA Ultra, the maximum input level is 6 dBm or about 0.5 volts. Turning on a 12 volt supply would result in a spike far higher than this, as would any accidental short-circuit or disconnection. Some voltage clamping is needed, and this will result in a bit of signal attenuation. I aimed for 10 dB attenuation in order to not decrease the sensitivity too much.

I did the output clamping in two stages. First stage consists of 25 ohm series resistance and four high-current DSS16 schottky diodes. These will clamp even large transients to 1 volt peak voltage on each line, while giving 6 dB of attenuation.

This is followed by the differential/common mode transformer setup. I selected SWB2010 RF transformer because it has the middle tap needed for common mode measurement. The models SWB1010 and SWB3010 are equally suitable, only difference is that the middle pin is cut shorter. A double-pole switch connects either the transformer secondary or the middle tap of the primary to the output circuit.

From the transformer the signal goes through 30 ohm resistance to a pair of BAT54 diodes that form the second clamp. These do not tolerate large current spikes, but act faster and with smaller dropout voltage. Because there is only about a volt coming into this stage, the current that passes through the 30 ohm resistor is less than 50 mA.

Despite these series resistors, the impedance at the capacitors should be kept at 50 ohms. This is accomplished by a few dummy load resistors connected between the signals and ground. These will dissipate as heat 90% of the input signal, leaving the 10% for the 10dB attenuated output connector. As an extra benefit, the LISN will now have a reasonably correct impedance even when no measurement instrument is connected.

Circuit board and enclosure

The PCB is a single-layer design, suitable for home etching or CNC routing. SMD parts are used, but nothing excessively tiny. The resistors are of 0603 type because that is what I commonly use, but 0805 would fit too.

I designed the circuit to be mounted to the TEM cell wall, with binding posts going to the device and banana jack connectors towards the power supply. A pair of M3 mounting holes provide the ground connection to the TEM cell shield through metal standoffs that I soldered to the copper.

The enclosure is 3D printed. It has a slider knob on the side that will hook into the switch on the PCB. By using shorter metal standoffs to attach the PCB, the same holes can also be used for mounting screws to hold down the lid.

Results

An ideal LISN would present 50 ohm impedance from each output terminal to ground at all measurement frequencies. In practice this is not feasible, so test standards specify either 5 µH or 50 µH series inductance to be used. For a 50 µH LISN, measurement range is usually 9 kHz to 100 MHz.

I measured the impedance using NanoVNA V2 and made a comparison against Tekbox TBL5016-1 .

The impedance on my build fluctuates quite a bit, but almost remains within the limits set by CISPR16 standard. I'm not sure what causes higher than expected impedance at the middle frequencies. The drop at high frequencies could be due to PCB stray capacitance.

Design files

KiCad design files are available on GitHub .

– Petteri Aimonen on 31.10.2023

Comment on this page

Usage experiences

Tue, 31 Oct 2023 00:00:00 +0000

Usage experiences

TEM cells are best suited for eliminating highest noise components early in prototype cycle, and for testing improvements after a failed EMC test. So far I have only tested prototypes, but the setup has already proven useful at spotting mistakes early.

Limits and background noise

These tests were performed without any advanced correlation to far-field measurements. A rule-of-thumb limit for TEM cell measurements is that anything over 40 dBµV (equals -67 dBm) is significant risk of failing in EMC tests.

Open TEM cells will catch some radio signals, even though they are not as sensitive in outside reception as actual antennas would be. The background noise in my office consists mainly of FM radio in 88 MHz to 108 MHz band and a nearby TETRA basestation at 390 MHz. Above 500 MHz there are more noise sources, including GSM base stations.

To be able to take the background noise sources into account in analysis, I have added gray background to them in TinySA-App .

Case 1: High conducted noise

This device is early in prototype cycle. The PCB integrates a isolated flyback DC-DC converter for power-over-ethernet operation. It also has a microcontroller and Ethernet interface. Early testing allows easy changes in PCB layout and component choices to eliminate EMC noise.

Emissions before modifications. blue: radiated noise, red: differential conducted noise, green: common-mode conducted noise

Zooming in on the fuzzy section below 100 MHz reveals that it is full of harmonics of the 300 kHz SMPS frequency. This noise is also very visible in the conducted measurements, indicating insufficient filtering on the power supply input.

To test the effect of filtering, I added a series ferrite bead and a 100 nF capacitor before that. Together with the existing decoupling capacitors, these form a low-pass pi-filter on the input supply. This should block especially the differential component of conducted emissions.

Comparison of emissions. blue: radiated noise before modifications, green: radiated noise after modifications, red: differential conducted noise

The modification eliminated practically all conducted emissions. It also significantly reduced the broadband radiated emissions.

There remain significant emission spikes at multiples of 50 MHz frequency. These correspond to the ethernet interface RMII reference clock signal. To address this, I'm adding a series termination resistor in next PCB revision.

Case 2: Excessive GPIO slew rate

A different device is further developed and may be undergoing EMC tests in near future. To reduce risk of failing the tests, a pretest look at any emissions is done.

Radiated emissions before modifications.

Overall the spectrum looks clean, but there is a significant spike at most multiples of 25 MHz. This corresponds to the clock rate of an external communication interface. The microcontroller used has a configurable GPIO slew rate, which is currently set to highest speed for the GPIOs involved. As a test, I reduced the slew rate to next slower setting:

Radiated emissions after software change.

This simple software change eliminated most of the emission spikes. Further testing and oscilloscope measurements are needed to verify that signal integrity on the communication bus is not affected. Thanks to the advance testing in TEM cell, these software changes can now be performed at leisure before the test laboratory visit.

– Petteri Aimonen on 31.10.2023

Comment on this page

Alphabet LEGO bricks

Sun, 05 Feb 2023 00:00:00 +0000

Alphabet LEGO bricks

LEGO bricks with alphabet are weirdly unavailable in the 1x1 size. There are DUPLO bricks and small round pieces with letters, but each with only one letter. Doing it that way needs a pretty big bunch of bricks to actually be able to spell something.

Instead, I decided to make 1x1 bricks that have one letter on each side. This way needs only one quarter the size of inventory of bricks on average. I bought a lot of 250 pieces 1x1 white bricks from eBay seller stein-experte . Now I just need to add the text.

Engraving

To engrave the bricks, I used the K40 laser cutter at my local hackerspace. The settings were 1000 mm/min speed and the smallest power the tube would run at.

An aligment guide made it easy to place a stack of 10 bricks for engraving. I would then incrementally rotate each stack and move it forwards to engrave each side.

Inking

The CO2 laser just melts the ABS and does not give much contrast in the markings. To make them more visible, I used a pen to color in the grooves. Excess ink can then be wiped out, leaving a well-defined letter.

Initially I tried an alcohol-based permanent marker. Unfortunately the alcohol caused tiny cracks in the ABS plastic, and the ink seeped into them. A water-based gel ballpoint pen worked much better, and after drying the marking seems durable enough.

Files

The Inkscape drawing of the DIY letter template is available here .

– Petteri Aimonen on 5.2.2023

Comment on this page