Upconverter for an SDR

Because the Lime SDR I purchased recently did not come anywhere near to expectations, because it was virtually deaf below 30MHz, I decided to use it as a tunable IF strip to a front-end converter. The last one of these I made was for the 2 meter band almost 60 years ago. I used my R206 communications receiver tuning 24 to 26MHz to listen to the band 144 to 146MHz. All the features of the R206 were usable over the 2 meter band including an FM demodulator which I later added to it, but all that was so long ago the only active components available were valves. The converter for the Lime is different. A tuning range of zero to 30MHz will be translated to a VHF band which I'm tentatively choosing to be 50 to 80MHz. This band is only lightly used so few signals will be present to upset matters. The upconverter will have four stages. First a low pass filter covering the band up to 30MHz, then an amplifier or buffer capable of providing some gain if I need it, a balanced mixer using four schottky diodes, and a crystal oscillator running at 50MHz. I'll be describing things in detail, and redesigning as necessary, as I proceed...

Above: The first low pass filter, and below, it's response. Note that the two 100 ohm resistors are not connected because the tracking generator and spectrum analyser I used to make the measurements each have 50 ohm terminations. The vertical scale is in dBm and the tracking generator output was set at -20dBm (which is 22mV). The horizontal scale is zero to 75MHz. The shape of the curve is approximate because of the low resolution bandwidths used (100KHz).

The first attempt was fairly successful, providing a good cut off (marker 1 is at 32.431MHz) and a reasonable ripple within the passband. I wasn't too happy with the out of band response and decided that this was probably due to inductance effects in the capacitor connections producing resonances so I decided to try smaller capacitors having less inductance. This turned out to be a bad idea as the pass band ripple deteriorated from 2dB max to over 7dB.

Above, the second design and below its response. Despite the physically smaller capacitors the flatness in the VHF region wasn't improved and the pass band ripple was poor and was prone to shorting turns as I tweaked it so I tried a third design using enamelled wire.

Before I changed the coils to avoid shorts between their turns I added a single transistor amplifier. This is a 2SC3355 transistor with a 22kohm auto-bias resistor and an RF choke as collector load. The emitter resistor was initially 910 ohms which gives about 12mA current with a 5 volt supply. The output waveform reflects the shape of the input and shows about 20dB of power gain. The in-band ripple is poor so I'll be refitting the original filter parts. Later I'll be doing away with the amplification provided by the transistor because it produced far too much signal for the SDR. Looking back, maybe I should have used an FET rather than a standard NPN transistor?

The in-band ripple was much worse with the modified design, going from a couple of dB to 7dB, so a third design was called for. This time I used 21SWG enamelled copper wire because I'd noticed it was too easy to get shorted turns with the BTC wire. I wound all the filter coils on a twist drill marked 4.8mm to give me a finished diameter of 5mm. I also swapped back to the earlier capacitors. Although the newer capacitors were much smaller the layout meant that lead lengths added unwanted inductance. Tests then showed the pass-band ripple was very good (around 2dB max) so I proceeded to design the remaining parts of the upconverter.

Above is the final version of the low pass filter giving attenuation of better than 50dB beyond 40MHz. I've also added a couple of BNC sockets to aid testing. Top right is a buffer amplifier using a small transistor, type 2SC3355 and below this a rough and ready double balanced mixer. The coax cable terminated in a 100 ohm resistor carries the 50MHz test input to the mixer (to be later replaced by a crystal oscillator). Using a signal generator for this allows me to work out the best power level to drive the mixer. Why did I use a 2SC3355? I looked through my collection of transistors and picked one suitable for easily soldering in place and with a really good fT.. the 2SC3355 has an fT of 6.5GHz. I tried this as a self-biased RF amplifier circuit using a 22kohm resistor and a 5 or 6 volt supply line (the red & blue wires which connect to a bench power supply) but during tests I changed to a fixed bias arrangement because I found this gave more linear amplification. The mixer uses a pair of double-diodes wired as a ring (not to be confused with a full wave bridge rectifier!) fed by a toroidal transformer adjacent to the 330 ohm resistor and an oscillator input toroid adjacent to the lower BNC socket. I'm using miniature 0.1uF DC blocking capacitors (the blue things). Initially I temporarily tried a large RF choke as the collector load for the RF amplifier but changed this for a 330 ohm resistor because the gain was peaking unduly at MF due to its resonance. The two toroids are each wound with enough turns to allow me to carry out initial testing but probably nowhere near right for best performance. In fact tests proved the upconverter is working sufficiently well to make measurable improvements.

At this point I'll cover some points raised by testing. I used two signal generators. A decent quality Wavetech which has an accurately defined output level in dBm or mV and an old valve-based equipment used in its day for general workshop testing. This is marked "Grayshaw Instruments SG50".This dates from 1954 when it was sold by it's Harpendon manufacturer for £6-19-6d. I also used a spectrum analyser and a 100MHz digital oscilloscope. The last item worked OK but I was unhappy about the readings (using a standard probe) once the RF was increased beyond around 20MHz. Below is a table showing apparently hopeless results from the RF amplifier. I imagine we're looking at several factors.. The effect of the RF choke, the falling-off of the scope readings due to its probe and the performance of the low pass filter as well as the (unregulated) gain of the transistor. At this point I removed the choke and fitted a 330 ohm resistor. (Note that once I'd changed the design of the balanced mixer the amplifier wasn't required.. in fact it completely swamped the SDR front end).

Input frequency-MHz  Input mVolts RMS  Output mVolts RMS  Gain dB  1  0.5  74  43  10  0.5  24  34  20  0.5  14  29  30  0.5  9  25

The next step was to see if the circuitry upconverted an input signal. I connected the old SG50 signal generator set to 20MHz to the input instead of the Wavetech, and used the latter as a 50MHz source for the mixer. Looking at the spectrum up to 100MHz I could see loads of signals. Because the balanced mixer is using temporary toroids I'm not altogether surprised to see this. For example the 50MHz signal is breaking through as is the input signal of 20MHz, but restricting the spectrum to the range 50 to 80 MHz I could see the upconverted 20MHz signal sitting at 70MHz so the thing is actually working. I could also see a few other strong signals, and by twiddling the tuning of the SG50 I found that one rogue signal was the 3rd harmonic of its 20MHz output. Not surprising because I'd noticed the sinewave on the scope from the SG50 was very distorted. I could also see a couple of other rogue signals but after fiddling with the transistor amplifier bias settings and reducing its standing current, these dropped in amplitude by 20dB. Next, I varied the amplitude of the Wavetech which was feeding the balanced mixer. The 70MHz output increased as the 50MHz signal was increased and I finally settled on 500mV. At this level I could see a 70MHz signal sitting at -38dBm with the noise floor at -84dBm. -38dBm is about 3mV in 50 ohms. Below is the circuit before making changes to reduce SDR overloading and to improve linearity.

The next step is to sort out the two toroids. The input toroid needs to work well across the range 50KHz to 30MHz and the output toroid across the range 50 to 80MHz. The latter's VHF coverage is important as it will reduce the effects of strong signals from the HF band from affecting the upconverted output. As a rule of thumb the inductance of the coils should be calculated to resonate at something less than the lowest frequency. The toroid also needs to be wound with the coils twisted together in a "trifilar" manner. To do this I cut three lengths of 30SWG enamelled copper wire, knotted them together and held the ends in a vice then knotted the other ends and held them in the chuck of a hand drill. I was then able to get a length of around a yard of decently twisted wires. I worked out that the input coil should be about 700uH and that of the output about 0.7uH. I fitted the new pair of toroids (seen below) and these worked OK. The next step is to carry out further tests although before this I'll need to add a 50MHz oscillator to the circuitry. This done... below is the upconverter with the internal 50MHz oscillator at the bottom right. I also added a metal screen around the low pass filter to aid its operation.

Initially, I constructed a 2N3819 FET oscillator but found it was tricky to set up as the crystal insisted on working at16.67MHz. Clearly it's designed to run in third overtone mode and wasn't happy in that circuit, so after redesign around a 2N5109 VHF transistor with a 100nH choke in its collector, tuned by a small preset trimmer, it operated close to 50MHz. After adding a 100pF capacitor from collector to ground, I found it worked on exactly 50MHz. I connected it to the double balanced mixer and wired the upconverter into my SDR Play. It worked after a fashion but loads too much signal, particularly from heavy local RF interference resulted in instability so I'll modify the amplifier following the filter to act as an impedance matcher with unity gain. Also, at some point I'll need to improve the mixer as significant oscillator signal is present at the output. I may add a simple notch filter to bring down its amplitude at the output socket. The new crystal oscillator has about 200mV RMS of 50MHz sinewave at the collector. This represents about -1dBm. DBMs can operate with say 0dBm (=220mV @ 50 ohms) so the level is probably just about right.

The next test was made after rewiring the buffer amplifier to work as an emitter follower with a 1kohm emitter resistor and around 10 volts on the collector. The output of the upconverter was fed to my SDR Play and the Wavetech connected as a signal source. I could see breakthrough from the 50MHz crystal oscillator sitting at -23dBm. A 1mV signal at 1MHz showed up correctly on 51MHz and 49MHz with both measuring -41dBm. Tuning to 2MHz pushed this down slightly to -42dBm and 3MHz to -44dBm. Still at 3MHz a reduction to 100uV resulted in -64dBm and 10uV to -84dBm. These measurements are what I'd hoped for as 20dB represents a reduction in input voltage of a tenth and 40dB a hundredth. The oscillator breakthrough was -23dBm which means it has a level at the SDR input of about 10mV. If the oscillator is pushing out 200mV, this means it's being reduced by 30dB although this could represent both breakthrough due to mixer unbalance plus direct pickup of the oscillator signal at the RF output. A narrow 50Mz notch filter in the RF output might help reduce the level? Further testing using the Wavetech showed that a 5uV signal at 20MHz gave -71dBm and a 30MHz signal gave -74dBm. Beyond this frequency of course the low pass filter attenuation comes into play. A 30MHz signal at 100uV showed up at 80MHz with a strength of -48dBm and increasing the frequency to 35MHz and 40MHz resulted in responses at 85MHz of -85dBm and 90MHz of -90dBm. The final test was at 30MHz when 500uV produced -38dBm and at 35MHz -70dBm. The low pass filter is therefore producing around 32dB of attenuation at +5MHz and 42dB at +10MHz which lines up nicely with the spectrum analyser curves above. Finally I set the SDR to 53MHz. A 1uV signal at 3MHz produced -103dBm and 0.5uV produced -109dBm. The noise baseline was -125dBm and using a 1KHz tone at 50% modulation I could just hear a 3uV signal. During the tests I was aware of masses of local electrical noise, some from fluorescent lights and some from a local surveillance camera operating over ethernet CAT5 cable. Further tweaking and tidying up of the circuitry is needed together with extra screening before I'll be happy with the results.

The next test was to add a notch filter to reduce the 50MHz breakthrough. A simple coil/capacitor reject circuit didn't produce much effect though (only 3dB), so I changed this to a small band-pass filter using three coils and some capacitors. Checking this using my spectrum analyser showed it was cutting off at 75MHz but changing the output capacitor shifted the cut-off to just beyond 80MHz. Out of band rejection was about 20dB.

Next I tried connecting the upconverter to my SDR Play. The 40 meter amateur band provided lots of CW signals but not easily measurable so I moved to checking Radio 4 on 198KHz. Using a long wire aerial gave me a signal strength of -19dBm with the SDR set at 198KHz. Swapping over to my Lime SDR gave me -27dBm for 198KHz and -6dBm for the same station at 50.198MHz. The Lime SDR includes internal circuitry which performs very poorly at frequencies below about 30MHz, but with an upconverter shifting 0-30MHz to 50-80MHz, this drawback is no longer important. I need to make some more measurements to see if I need to improve the filtering and redesign operation to use 5 volts from a USB socket rather than having to rely on a bench supply. The diode ring needs to be tidied up as it's dangling in the wiring at present. A more robust circuit might use 1N5711 diodes and these could be fitted to balance out the circuit and reduce oscillator breakthrough which currently may be having an effect on AGC? Another feature might be a small relay to enable straight-through aerial connection. This could be operated by connection of upconverter power, so unplugging a USB power connector would switch the upconverter relay off, thus connecting together the input and output sockets. Replacing the rather fragile set of diodes looks like it's going to be straightforward by replacing the pair of tiny chips with a single quad ring. I found a suitable type, an HSMS-2829 which seems to be ideal. I also plan to replace the ferrite rings. One with a VHF range and the other specified for MF/HF. At the moment, both my toroids are from my junk box and are coloured yellow. The upconverter does work reasonably well but I'd like to improve the performance, especially in terms of reducing spurious output signals.

This is the first version of the diode ring that I used. It consists of a two BAT74 Schottky diodes in SOT23 packaging. I happened to have some of these left over from a lift repair. They were awkward to solder and not balanced very well so allowed too much of the 50MHz oscillator to get through to the RF output. Click to see the spec..

I got around to ordering some better diodes which I fitted in place of the BAT74 diodes. This is a Schottky diode ring with the part number HSMS-2829-BLKG and should be better balanced. Click to see the spec.. The input toroid is wound with 30SWG enamelled wire. Three lengths are twisted together with about ten twists per inch. Two lengths are connected in series with the end of one soldered to the start of the second. The junction is therefore the centre tap. The third wire is the primary winding. This means that the primary has half the turns of the secondary, thus giving a step up ratio of 1:4. If the primary is designed for 50 ohms the secondary will be 200 ohms. I wound 30 turns (this number is dependent on the type of ferrite). The output toroid is similar in construction but has fewer turns to reduce its efficiency below the 50MHz oscillator frequency. I wound 8 turns using the same wire and twist as the input toroid.

For anyone wishing to duplicate the upconverter the coil winding details and capacitor values are given below under the picture of the latest version although as you can see, some parts are not used and it needs tidying up.

The low pass filter at the top of the picture has four coils each having 14 turns wound on a 3.8mm former to produce a diameter of 5mm and a length of 10mm. The input is coupled to the BNC connector via a 100nF capacitor. The relatively high value is to reduce the input resistance to long wave signals. The input of the first coil is decoupled to chassis via a 200 pF capacitor and then the output of the first and 2nd coils by 300pF capacitors. The output of the fourth coil has a 200pF capacitor to chassis. The filter output connects via a 100nF capacitor to the base of a transistor. At this point in development, the base was self-biased by a 22kohm resistor. The bias voltage results in current being drawn through the 91ohm emitter resistor. The collector is tied to the transistor supply voltage which is limited to around 12 volts by the temporary collection of resistors you can see top right. RF output which is substantially at the same voltage as the incoming signal is coupled via a 100nF capacitor to the input toroid of the double balanced mixer. The 50MHz crystal oscillator circuit is very simple and uses a small coil tuned by a small trimmer capacitor to run the crystal on its 3rd overtone. This is temporarily connected via a 100nF and 100 ohm resistor in series with the output toroid. You can see the diode ring chip between the two toroids. The output toroid connects to a simple bandpass filter covering about 50 to 80MHz. There are three coils. The input coil has 12 turns and connects via a 22pF capacitor to the shunt coil to chassis of 4 turns in parallel with about 180pF of capacitance. The output coil and capacitor are the same as the input coil. Again, everthing needs tidying up and the larger coils should really be at right angles.

When I'm happy with the final result I intend to solder more tin over the low pass filter etc to reduce pickup from local interference sources.

Below is a picture of the converter output which will need a little explanation. I'm using a tracking generator to produce the scan which goes from 100KHz to 100MHz. The left end shows the effect of the 30MHz low pass filter superimposed on which is a 25MHz -40dBm output from the crystal oscillator (I'm not sure that this should be present if the oscillator is working correctly). Next is an area of attenuation which extends to 50MHz where you can see the 50MHz crystal oscillator signal sitting at about -32dBm. The "marker" is resting on the baseline of the 50 to 80MHz band pass filter.This filter is relatively simple and has a sag of 5dB. Then there's an area of attenuation (at -80dBm) from about 82MHz designed to reduce the effects of local FM broadcasts.

Below is a narrower scan from 50 to 80MHz (=0 to 30MHz) where you can see the variation in the bandpass filter characteristics. This shows a 2MHz signal from my signal generator which is Tee'd into the tracking generator output. The horizontal lines are 7dB apart, so the ripple is 7dB. As the signal generator is tuned upwards the spike moves to the right where 30MHz is upconverted to 80MHz.

Here's a further scan showing a signal of about 120KHz sitting on a scan from 50 to 51MHz (=0 to 1MHz)

Below is a scan of the low pass filter taken from the emitter of the transistor buffer after adding a 330nF decoupling capacitor at the collector followed by an increased scan where you can see the small reverse leakage from the 50MHz crystal oscillator. In these scans I'm using a signal probe using a 1Mohm series resistor connected to a 10Mohm input circuit. This basically adds insignificant loading to the measurement points.

This next scan is made at the 50 ohm output connector. This shows 50MHz breakthrough with a small amount of its second harmonic at 100MHz. You can see the shape of the bandpass filter which rolls off at Marker 1 at 80MHz corresponding to an upconverted frequency of 30MHz. By adding more filter sections the attenuation could be increased to around 30dB.

Here's a screenshot of the Lime SDR with the upconverter and tuned to about 57.160MHz corresponding to the 40 meter band. The receiver is tuned to a lower sideband signal registering -81dBm. I'm using an 80m inverted V so signals are not particularly good but the level of local interference is much lower than from a long wire.

Below is the latest circuit for the G3PIY upconverter Note that the length of coils L1 to L5 is 15mm, L5 & L6 are 12mm and L7 is 6mm. L8 is 10 turns on a 3mm former. The oscillator bias resistor is now 47kohm not 120kohm. L8 trimmer is now a 22pF fixed capacitor. The buffer emitter resistor is now 43ohms. Most changes were done to reduce the supply voltage requirement to 5 volts. The toroids are wound to suit their material. Mine are coloured yellow and are a bit less than 5mm diameter and 2.5mm thick. See later for the reason for the extra 50pF.

I decided to tidy up the upconverter and see how it performs. Across the whole range to 30MHz I can easily resolve 5 microvolts from my signal generator, in fact you can see a blip down to less than 1 microvolt. The blip vanishes, as it should at 31MHz and above. Response is much the same down through MF and LF . Plugging in a convenient dipole, actually cut for 80m, results in strong reception down to ELF. I moved the receiver to my main computer and tried a frame aerial which I use for my SDR Play. This aerial gave me reception of our local long wave station with a maximum indicated signal level of -3dBm. This was achieved by fiddling with gain settings and of course these are certainly not optimum. The very best I can see with an SDR Play is around -22dBm.

Above, slightly tidier. I modified it to work on 5 volts by substituting an RF choke for the crystal oscillator collector resistor but the output from the oscillator dropped and reduced the mixer performance so I run the converter on 12 volts instead. I added a couple of ceramic standoffs to make the construction more rigid and increased the oscillator coupling capacitor to 330nF from 100nF and removed the 100ohm series resistor. The oscillator bias resistor is now 47kohm and I soldered the crystal to the tinplate. I also reduced the buffer amplifier emitter resistor to 43 ohms from 1kohm. These mods were made so I could reduce the operating voltage to 5 volts. It actually works on 4 volts but I found that pushing this up to 12 volts increased overall gain.

Above is the upconverter rebuilt into a small diecast box. I've added a screen along the centre line and screens between the coils in the low pass filter and placed a metal screen between the crystal oscillator and the input toroid to the double balanced mixer. Also, I've put a screen between the coils in the band-pass filter. I used veroboard to minimise chassis currents between the input and output circuits. The veroboard was cut to fit the box, then assembled and fitted in place. Once this was done I soldered the BNC connectors to the filters. I haven't checked yet to see if any improvements are noticeable however, below are two screenshots using the Lime SDR, one of the 30 metre band using a poor antenna. The receiver is tuned to 9.570MHz. See further on to explain the reason for the arrowed note. The second is the band from zero frequency to a little above BBC Radio 4 on 198KHz. A couple of time signals can be seen in the waterfall. The aerial was a short length of wire local to the computer with associated chopper power supplies whose rough signals can be seen. Oscillator breakthrough is at 50MHz.

Below is the same spectrum for the Lime SDR, but without the upconverter and using the same wire aerial as the picture above. Radio 4 is unreadable and the only signals visible appear to be interference from chopper power supplies. One noise source is producing spikes at 80KHz, 160KHz and 240KHz with the second producing humps of noise at roughly every 11KHz.

Once I'd familiarised myself with the operation of the upconverter I decided to check it on my spectrum analyser. By feeding the input with the tracking generator and monitoring the output I could see the overall response of the converter (picture will follow soon). The results are a little puzzling because several things are going on simultaneously. The input filter reduces the overall response above about 35MHz and the output filter reduces the response above 80MHz. The level of the crystal oscillator is quite significant but should only interfere with very low frequency signals. Having checked the response I injected an RF test signal of 10MHz into the input and this duly appeared as a 60MHz output. Initially I was puzzled because the size of the upconverted signal was switching between two values but this was corrected by turning off the tracking generator. As with most home-brew projects you're left with the doubt that it's working as best as it can. I'm reminded of the radio designer a Mr Scott-Taggart who produced receivers in the 1930s where almost every passive component was variable resulting in a multitude of front panel controls. I can also point you you to the DST100 receiver designed for absolutely the maximum performance in reading "almost" unreadably weak transmissions from U-boats. To this end I then checked the sensitivity of the converter to the effect of adding 50pF extra capacitance at various points in the circuit and found little of significance except around the mixer coils. I discovered that by touching 50pF across the primary winding of T2 pushed the converter output up by over 10dB. Not only did the ouput rise in amplitude but the "width" of the signal reduced, looking much cleaner so I soldered the 50pF capacitor in place (see circuit diagram above for its location). Unfortunately, the extra output completely swamped both the SDR Play and the Lime. The effect was to produce a comb of spikes spaced by about 17KHz around the 50MHz breakthrough signal and any tuned stations so I snipped off the capacitor and all was well again. By moving the RF test signal up and down from 1MHz to 30MHz I found the corresponding output 51MHz to 80MHz followed as expected. The output remained pretty well constant from 30MHz down to less than 1MHz when the output increased (which is no bad thing as it will improve MF and LF broadcast signals). I then carried out a test on the effect of the power supply. As I'd planned, the crystal oscillator worked down to 4 volts before cutting out. Overall gain did improve though as I increased the voltage (mainly due to the increasing level of the 50MHz mixer local oscillator). To check nothing untoward was happening as the supply voltage increased I checked the crystal oscillator harmonics to check for unwanted spurii. See the table below which shows the harmonics not worsening and even reducing slightly as the supply voltage is increased, so operating from 12 volts is fine.

Supply Voltage 50MHz level = f0  100MHz level = f2  ratio f0/f2  5  -46.18dBm  -61.74dBm  15.56dB  8  -40.90dBm  -57.16dBm  16.26dB  12  -36.06dBm  -53.47dBm  17.41dB

I carried out some tests using the Lime and found the following looking at the 50MHz breakthrough signal from the upconverter...  Input  RX1 W  RX1 L RX1 H RX2 W RX2 L RX2 H  50MHz  -40dBm  -51dBm  -39dBm  -40dBm  -42dBm  -38dBm

I then tuned the Lime to the 40m amateur band, tuning to 57.000MHz to 57.200MHz and compared the results across the different inputs. All the inputs performed much the same but I needed to individually adjust the overall gain setting on each input to around -62dB give or take to prevent overloading. Removing the upconverter and repeating the test showed that only RX1 L was usable but it was considerably worse than the upconverter results. Tuning down to Long Waves using the upconverter (50.000MHz to 50.500MHz) I found results were pretty good but as gain was increased and a tuned station became stronger the breakthrough 50MHz signal became modulated from the received station. I discovered the reason for the 50MHz signal being modulated by a tuned signal audio. I'd fitted the final assembly into a diecast box with a pair of BNC sockets side by side at one end but I'd forgotten to connect any of the ground connections from the PCB to the outer case. The result was earth currents were getting into the output cable and modulating any signals feeding the output cable. I experimented by shorting various points on the metal shielding to the case and found a solid connection at the screen dividing the two BNC connectors almost completely removed modulation. I'm sure I could improve things further by grounding other sections of screening to the case also (See picture above). Later I found that there was still some residual amplitude modulation of the 50MHz crystal by incoming strong signals. I also found some hum. By tuning the Lime SDR to exactly 50MHz, listening to the receiver and prodding around, it seemed to me that earth currents in the converter were responsible. I tried shorting various earth points to the diecast box and found varying results. The best result was found by connecting the base of the RF transistor to chassis via a 3.3pF capacitor. This eliminated the annoying hum on the 50MHz signal and which was being transferred to one or two broadcast signals. This capacitor did also dramatically reduce the modulation from the strongest signal entering via the aerial (this was primarily Radio 4 on 198KHz) and was visible in the SDR waterfall. The remaining modulation is now reduced to a very low level. A secondary improvement is that the receiver noise baseline is less and a lot flatter. Typically 50MHz leakage is -44dBm whilst Radio 4 is -61dBm with the Lime wideband input and gain set at 50dB. Then, I decided to compare the results using the SDR Play receiver. I found the 50MHz signal stood at -41dBm with Radio 4 at -56dBm which are pretty close to the Lime results. The SDR Play RF gain was set to Max and IF gain at -45dB (Auto). I then checked reception of Radio 4 on my Andrus. This registered -43dBm with RF gain set to 0dB, -42dBm at -12dB and -24dB and -49dBm at -30dB. Build an FM Rejection Filter to reduce breakthrough Below is a first attempt to build a filter designed to block FM broadcast signals breaking through in shortwave bands.

The filter circuitry is fairly obvious from the picture above, three bare coils between input and output connectors using tiny coils with 100pF shunt capacitors and four enamelled coils with series trimmers as reject circuits. The bare coils are around 25nH tuned by 100pF capacitors and determine the cut-off of the filter. These coils were initially slightly larger resulting in a sharp cut-off at 58MHz but now they're smaller the cut-off is around 80MHz. The reject coils are stagger tuned to provide a wide reject band up to 108MHz. Tuning it proved very difficult because I plugged in a coax lead that wasn't connected to the spectrum analyser because of extreme untidiness on my work bench. The correct lead feeding the analyser was picking up an extremely weak signal displayed by using its "auto" button. The pickup resulted in a picture which changed as the trimmers were adjusted but then I noticed the vertical scale was in small fractions of a dB. The penny dropped and a rummage around the bench revealed the wrong lead was being used. After discovering my mistake I switched the leads over and found the filter worked perfectly (see the waveform below). The plan is to rebuild the filter in a diecast box now that I understand the interaction of coil sizes and capacitances. All the coils are wound on a 3mm drill. Insertion loss is about 3dB.

Having tested the first attempt I wasn't very happy with its performance so I made another but using heavier wire to improve the Q of the coils. This is shown below. Its made on a small piece of tin cut from the lid of a box of Qualty Street chocolates. I used 50pF capacitors because I have a large quantity of these, and 18SWG tinned copper wire. The trimmers are something like 3-30pF. There are 5 coils in series (4.5 turns on 5mm) with 3 shunt coils (12 turns on 5mm). The performance is shown below the picture. The design probably has a fancy name?

Marker 1 is set at 80MHz and Marker 2 at 109MHz. Marker 3 is at 118MHz. The attenuation over the FM broadcast band is about 50dB. The tracking generator is set at -20dBm and some losses will be due to cables and connectors. How did it work in practice but with the filter as shown and unscreened? Classic FM dropped from -77dBm to -107dBm so in practice I got an attenuation of 30dB. That station went from perfectly clear and very strong to scratchy and very weak. Unfortunately there were consequences. Testing the filter directly into the Lime resulted in masses of extraneous noise spikes and lots of out of band signals. I moved the filter outside into the antenna feed line and this removed most of the local electrical noise and revealed the out of band signals. Essentially one needs not a reject filter but a band pass filter centred on your band of interest. That might be my next step... a low pass filter for say the AM broadcast band and a bandpass filter for each specific band in which I'm interested. When the filter is properly enclosed the curve should be much the same shape but stray pickup of RF will be reduced. Below is a test carried out with the screened filter and the SDR Play tuned to 88.5MHz. First with the filter in-line.

Now, the filter removed but with the SDR settings untouched.

Now with the settings adjusted to position the noise floor.

What do we see? With the filter installed the signal has much the same amplitude as peaks of interference with the noise floor at higher frequencies raised slightly, perhaps from a strong signal. Without the filter.. The noise floor is raised significantly from other and in-band signals and finally the last picture shows the noise floor at the same level but pushed down the display to give a better picture. The received signal is -101dBm with the filter and -65dBm without.. indicating a rejection at 88.5MHz of 36dB. At this point I haven't tested the filter on the spectrum analyser after fitting it into its screened case so this figure might be improved a little. See below.

The experimental filter in its case. For convenience I'm using F connectors on flying leads rather than connectors fitted to the case. The holes are for trimmer adjustment.. two in the top and a third at the side.

During tests using my SDR Play I saw virtually the whole spectrum full of intermittent spikes. These were grouped in sets of half a dozen. I recalled I'd seen this problem a few days ago and, guessing the cause, tuned to the Citizens Band, which in the UK occupies 40 channels from 27.60125/27.99125 MHz (Channel 1 to 40 ). The CB craze has progressively faded but there are still operators around, some of whom use large aerials and very high power (both illegal). Unfortunately local CB and an SDR are not entirely compatible due to overloading of the SDR untuned input. I inserted a variable attenuator in the aerial feed and found that by adding 20dB of attenuation the mass of 27MHz spikes disappeared and I could see the true frequency of the CB operator. The breakthrough on other bands was now much reduced, however all the other signals at the aerial are also degraded by 20dB. The answer has to be a reject filter for the band centred on 27.7MHz. Not too difficult to make because its bandwidth needs to be only about 400KHz. Below a plot of the cased Broadcast FM reject filter, Markers..1=66MHz, 2=85MHz, 3=108MHz, 4=120MHz. Ideally, one of the shunt coils needs compressing slightly to reduce the air-band attenuation. The tracking generator has an output of -20dBm so attenuation below 40MHz is virtually zero so the filter is ideal for HF reception.

Now to transmit... The BIG question...How does the Lime SDR work when used as a driver for a transmitter?

See the Lime Transmit experiments

Return to Reception