Noise Source

Below is shown the circuit of a typical noise source on which my version is based.


 At this stage I plan to use a BGM1013,115 in place of the BGA2817 and a 6 volt battery with appropriate changes eg a different zener diode.

I should explain the purpose of building this noise source as it's not intended to compete with expensive commercial items, but merely as a project with which to familiarise myself with the interesting topic of noise in radio receivers and perhaps to see how it compares with a tracking generator for helping with RF filter design. That said, read on to hear how I built it then attempted various design improvements.

Click here or scroll to end of this page to see the final circuit diagram.

To eliminate local RF pickup it will be put in a diecast box and the construction will be on a piece of tin bent under the BNC output socket securing nut with the chips mounted on grounding wires and the battery supply decoupling capacitor(s), The zener diode and various resistors will be soldered directly to the tin. The power supply will be three or four AA cells mounted in a holder within the diecast box with a tiny LED and on/off switch. The choice of the BGM device was partly because it can accommodate 6 volts whereas the BGA devices are limited to a 4.5 volt battery supply as they can only handle 5 volts, however I'll try and extend battery life by balancing the supply voltage with total current drain. These types of chips sink current depending on the amount of signal at their input. As the BGM chip has a higher gain than the BGA, maybe one chip will be enough?

Up to now, I've not considered a suitable zener diode. In the circuit above you can see the front part of the circuit has the title "Avalanche" because a zener diode can have a couple of basic designs and the avalanche type is said to have a better noise characteristic for our noise source. Many years ago, back in the 60s and 70s our factory made noise sources for use in cryptographic equipments and their design was based on zener diode breakdown noise, so relatively early zener diodes would have performed quite adequately. The noise source I'm planning will use a 6 volt battery so I'm looking for a zener diode having a rating of something less than 6 volts. A check on my junk box reveals 11 zener diodes rated below 5 volts including some ancient obscure types. I also have a good selection of 5.1 volt and over a hundred others. The plan is to try a selection and if none below 5 volt are good enough, raise the battery voltage to 12 volts and try others. If any specific types are significantly better I can order a lower voltage rating. This will also apply to TVS diodes, as I only have a few rated at below 9.1 volts. The diagram above shows 100nF capacitors in parallel with "uF" implying you need fairly low impedance coupling for lower frequencies. For RF work 100nF to 1uF should be OK. The output matching network shows a PI configuration with a 36 ohm series resistor and a couple of 360 ohm shunt resistors. Another solution is to use 3 AA cells plus a further 5 AA cells for the zener.

 I ordered the following parts including an SR2.8 device with an interesting spec
 BGM1013  5  RF amp, 35.5dB gain SOT363
 Diecast box  1  112 x 62 x 31mm (later changed to 121.2 x66 x 35.3mm)
 BNC connector  1  bulkhead screw fitting
 Battery holder  2  4 x AA cells (later changed to 6 x AA cells)
 SR2.8.TCT  2  TVS array, SOT143 low cap with 3V punch through (not used)

 The amplifier chip I've chosen needs a voltage at its output pin which is supplied via a choke. In the application notes this is shown as 100nH which at 100MHz represents an impedance of about 60 ohms. If the noise source is to work well down to say 100KHz this choke needs to be 100uH and at say 465KHz around 20uH. This means that I'll need to experiment with the output crcuit, perhaps using a load resistor in place of, or in addition to, the choke of say 60 ohms. The value of the resistor will result in a voltage drop at the output pin, but as the output current isn't specified some experimentation will be necessary. Also needing checking is the manufacturers "K factor" which appears to be concerned with instability, so where it drops to less than unity at 100MHz might result in a problem getting the amplification down to 100KHz?

The parts arrived but I'd forgotten to check the dimensions of the amplifier chips which looked fine in the suppliers catalogue but almost invisible when I looked in the packet (the body is 2mm x 1.25mm and the 3 legs on each side spaced 0.65mm apart). Back to the drawing board so I can work out a layout because I'll need to use mostly chip resistors and chip capacitors to balance the construction. Also, remembering the design of those TRF receivers using those fancy high gain tetrodes with a metal screen between input and output, I've decided to use a similar technique with the noise source to prevent it oscillating. Below.. a rough idea of the layout. The shaded areas are grounding wires soldered to the baseplate. Practical difficulties (mainly eyesight) meant I had to deviate from this plan.


 It looked easier than it actually was. First I found my magnifying desk lamp wasn't powerful enough to even see the dot on the chip marking pin1. I have a very high power lens which did the job, then I must have breathed too heavily and the chip vanished. Fortunately I'd noticed it was magnetic and I eventually found it on the floor together with umpteen ends of wire etc. I first soldered a pair of 100nF chip capacitors to the tin sheet, one either side of the central shield, then used one of these on which to mount A1 by soldering Pin 1, the power supply leg, Then I soldered ground wires to pins 2, 4 and 5 and soldered another 100nF near the output (Pin 3) and a 100 ohm chip resistor between this and the output, together with a thin wire whose end passed through the centre shield hole. Next I fitted a BZX79-C3V9 (3.9 volt zener diode) and a 1.2Kohm load resistor, connecting their junction to a 100nF capacitor to A1 Pin 6. Before adding A2 I then tested the A1 circuit by terminating this into a 1Kohm resistor via another 100nF capacitor. After setting the voltage to 5 volts and limiting the current, then gradually increasing the trip I found the current measured around 39mA.

Next I connected my spectrum analyser to see the results. I could see some large RF spikes even before connecting to the noise source. The main spike was around 600MHz and this was accompanied by a strong second harmonic at 1.2GHz. Clearly the high amplifier gain was resulting in feedback and, by intially adding a metal screen over A1, then trying different cables to the SA I settled on a BNC cable with a very short end which I soldered to the noise source output. By experimenting I found the noise introduced by the zener diode was very low (no more than a 1 or 2dBm increase over the minus 90dBm baseline) and oscillation occurred very easily. I swapped the 3.9 volt zener for a forward biased OA90 diode and found this was useless because I got an untameable 1.2GHz output signal at around 2dBm with zero noise. Next I tried a chip zener diode rated at 5.1 volts (see below for details). I increased the supply voltage to around 6 volts and found this zener did produce some noise, but with a very lumpy shape and still prone to oscillation. I added an extra metal screen which helped settle the instability.

As the output was still running into a 1.2Kohm load, I changed this into a pad comprising three 100 ohm resistors. This reduced the tendency to oscillate. I could now see RF broadcast pickup so I added a 330uF capacitor at the power input and passed the supply leads around a ferrite ring. The results were much improved and stable enough to see that significant (wanted) noise was being produced when the power was applied. Here are a set of pictures showing spans of 30, 100 and 1500MHz without power, then with the noise source switched on.



 Above, the span is zero to 30MHz with the noise output 24dB above base level.



 Above, the span is zero to 100MHz with the noise output 28dB above base level.



 Above, the span is zero to 1.5GHz with the noise output dropping rapidly to base level. Because RBW/VBW has been increased tenfold for the greater span, the start of the trace is not as well defined (about 10dB higher) as the previous pictures, but you can clearly see the noise isn't nearly as great as in the lower frequency ranges. Note: One of the problems during testing was hand capacity and at this point the circuitry was mounted on a small tin sheet and later, even with this in place inside the diecast box, the circuit was prone to instability, however, once the lid was screwed in place it was fine.

The noise source isn't complete yet as A2 hasn't been fitted but it's clear that building it in this fashion is not easy and it's quite possible that the addition of the second amplifier may make the project impossible to complete unless the overall gain is reduced to prevent instability. Stray RF pickup is also a problem and will remain so until the circuit is fully screened, however the choice of zener looks OK. This is a tiny chip device type MMSZ5V1T1G. The next step is to vary the zener load resistor to see if the noise output can be improved. It's also important that the range below 1MHz is reasonably flat because that range is most useful for IF amplifier alignment. Below.. not much to see.



The supplier's picture of the BGM1013, 115 

2mm x 1.25mm x 1mm

Right, sitting on top of a 6BA screw.

Below, a view of testing the first stage A1, then adding the second stage A2 is shown in the following pictures.







 Steps in fitting the second amplifier chip A2, are very similar to fitting the first chip, now under the metal cover to prevent RF feedback. Note the output wire from A1.

Top left.. securing pin 1 of the chip to the 100nF power supply decoupling capacitor

Top right.. adding the 100 ohm output load resistor between Pin 3 and a second 100nF power decoupling capacitor

Left.. adding the 100nF output coupling capacitor and ground wires to Pins 2, 3 (input) and 4 (output) of the chip.

Later, the 50 ohm RF output load (two 100 ohm resistors in parallel to provide mechanical stability) and power supply connections were added.

You'll realise that two soldering iron bits (and temperatures) are required.. one for use on the tin sheet and the second for soldering parts.

 Again, the noise source was tried outside its screening box and I found it was prone to oscillating at between 600 and 1000MHz. Below are scans at a span of 1MHz with the noise source power turned off and then turned on. The centre is at 500KHz, and the area around this is about -90dBm (S6, OFF) then -40dBm (better than S9+30dB, ON) giving a noise level of 50dB above baseline which is about double that for the first stage. If you look carefully at the first scan below you can see spikes indicating the presence of Radio 4 at 198KHz and Smooth Radio on 828KHz. A few of the other spikes are from local interference sources. The falling off to the left of the second curve is where the impedance of the 100nF coupling capacitors is within the same order as the output impedance of 50 ohms (ie. 100nF = 16 ohms at 100KHz and 160 ohms at 10KHz). The value of the capacitors could be increased from 100nF but this may have introduced annoying stray inductance at higher frequencies. The capacitors I'm using are from a batch I bought a few years ago for a few pence each and are physically ideal for this method of construction. I guess the use of suitable 470nF capacitors throughout would improve the LF end of the noise spectrum but this remains to be proven.



 The final breadboard test was to plug the noise source into a general coverage receiver (my Kenwood R2000) and I found the noise level was a pretty constant S9 +30dB (as predicted in the above scan) from 100KHz to 30MHz. The supply was 6.4 volts and the current consumed about 79mA which makes the zener current a little over 1mA. The next step is to fit the chassis into the diecast box, but this proved difficult because of the size of the AA cells. AAA cells would have made the task a lot easier.

I found a suitable larger box (Hammond 1590N1, approx 121mm x 65mm x 40mm) which will just accommodate the noise source plus twin 4 x AA battery holders to take 6 NiMH rechargeables and I'll fit a 2.1mm DC socket with a suitable series resistor for feeding a charging voltage. The surplus green box will be re-assigned for a new project.

While I'm waiting for the replacement box so I can finish testing the noise source...I wondered what I'd see on an oscilloscope. The power measurement by the spectrum analyser at any specific frequency (or range of frequencies) is about -40dBm. But that checks the level across its internal 50 ohm input circuit, and at the moment I'm using two 100 ohm resistors in parallel at the noise source output which is 50 ohms. Surely half the output power is dissipated at the two 100 ohm resistors and half in the spectrum analyser, so the total power being developed is twice that shown in the scan above? I recall the handbook for my TF2008 signal generator explains at length that the output voltage shown on the dial is out by a factor of two when connected to a 50 ohm input, but accurate if recorded on an oscilloscope using a high impedance probe.

Minus 40dBm represents about 2.2mV RMS or 6mV pp across the pair of 100 ohm resistors (=50 ohms) so this voltage should be visible on an oscilloscope. My GDS1102U has a maximum sensitivity of 2mV RMS per vertical division, and should therefore produce a 3.3cm pp display and if I use a 50 ohm Tee to connect the spectrum analyser the voltage at the Tee should be 1.5mV RMS or 4.5mV pp. In fact this all matters very little. Half the voltage is -3dB so the spectrum analyser, when it tells me it can see -40dBm across 50ohms is really saying the ouput is -37dBm because its input is shunted by my noise source output of 50 ohms. To read the true output I could change the pair of 100 ohm parallel resistors into say a single 500 ohm resistor. The resulting impedance would then be 500 and 50 ohms in parallel or 45 ohms and about half a dB instead of a 3dB error. The reading on the SA will now be closer to the noise source output when its terminated into 50 ohm, but if I now disconnect the SA and instead connect an oscilloscope probe, the display will now show a larger trace than before because the output will be across 500 ohms instead of 50 ohms. Assuming the same power is being produced this will mean (V1 x V1)/R1= (V2 X V2)/R2. Assuming the power output is -37dBm across 50 ohms, V1 = 3.1mV and -37dBm across 500 ohms, V2= 9.8mV. So the trace will increase from 3.1mV when the output was terminated in 50 ohms to 9.8mV with the 500 ohm termination. The problem with this is that when plugged into a receiver having an indeterminate input impedance the noise level will be equally indeterminate so leaving the terminator at 50 ohms will at least guarantee the noise input is between -37 and -40dBm at 50 ohms. In the end I decided to use a 100 ohm load within the box as this compromise would better tame the circuit and prevent it oscillating.


 I worked out the best layout for the batteries in the new diecast box and marked out drillings for the tin chassis, a small on/off switch and an LED. I have a collection of LEDs so looked through the box and tried a few. A yellow one in a holder drew 2.5mA at 2.4 volts. I tried another and it failed to light so rather than unsolder its connections I temporarily switched over the red and black plugs at the power supply (a really bad idea). This also failed to light te LED so I tried a third which was fine, drawing only 1mA at around 2.2 volts. This had a formed shoulder so was suitable without a special bulkhead fitting. Next I drilled three holes in the end of the diecast box and ftted the new LED, the miniature switch and the BNC connector. Behind this connector I fitted a tin sheet with a shoulder to which I'll solder the noise source tin chassis suitably trimmed to minimise the space required.

A tip here in mounting parts to the diecast box... to ensure a tight fit for the LED I drilled a hole slightly smaller than the LED diameter then used a scissor blade to carefully enlarge the hole for a tight interference fit. Diecast material is an excellent material with which to work except for soldering. Tin sheet is ideal for soldering and an excellent material for quickly rigging up a circuit. Once I was happy with the circuit I snipped off excess tin sheet, cut a second piece of tin, in which I drilled a hole for the BNC connector then bent it to fit the end of the box. I then soldered to chassis to this. I found I had to ground the opposite end of the chassis so drilled a hole through the chassis and box and secured the two with a 6BA screw. That completely tamed remaining tendency to oscillate.

Before going further and powering the noise source I moved the amplifier supply connections (two each to the pair of ampifiers) so that they are fed via a series 180 ohm resistor so that the amplifier 6 volt max supply voltage wouldn't be exceeded by the new 6-cell battery. Turning on I found a problem. The new assembly drew far too much current (upwards of 400mA) and a monitor receiver showed complete lack of noise. After experimenting a little I discovered the zener diode was sitting at half a volt and each amplifier independently drew well over 100mA. Now, during tests I'd noticed that my PSU had developed a fault... as the rotary control is turned to increase the voltage it would sometimes add a few volts to the output instead of allowing this to increment by 100mV per step. Had I inadvertently destroyed the amplifiers? I gave up for the day, intending to use two spare amplifier chips but then thought back to the zener diode. This surely couldn't fail as it's fed via 1.3K ohms. Then the penny dropped... I'd switched over the leads to the PSU when checking LEDs so red was negative and black positive and I'd forgotten to put them back. I switched leads around, set the output to 5 volts and 100mA, turned on the monitor receiver and switched on.... the receiver roared into life showing a steady S9+30dB right across the range from 200KHz to 30MHz, and dipping only slightly down to 100KHz. The current being drawn was very low so I set the output to 7.2 volts (the intended new battery voltage) and it drew 22mA. Dropping the voltage below the zener voltage turned off the noise output which is what I'd expect. The received noise hardly changed with supply voltage above 5.5 volts so I can now confidently finish assembly work.


 Above are pictures of the completed noise source, or at least the provisional model, as I'm not sure about the final output level. With a reduced power consumption of 22mA the output has dropped from 30dB to around 20dB above the minus 90dBm baseline across the shortwave band to 30MHz. No doubt this will help to reduce any tendency to oscillate as was the case when it was drawing close to 80mA. My previous diecast box project used alkaline cells which need replacing every so often, so I've now used rechargeable cells, adding a charging circuit based on a convenient 12 volt power supply. This feeds the battery via a 47 ohm resistor to provide 100mA charging current but I decided to add an orange charging LED. This proved slightly tricky to arrange circuitwise so I fitted the LED in series with the 47 ohm resistor. This reduced my initial 100mA charging current, but there's 2.8 volts across the resistor so that charging current works out at 2.8/47 = 59mA for 1.2 volt cells which is OK and to fully charge the 1300mAh battery will take around 12 to 24 hours. I used a handy single pole 2-way on/off switch so that the noise source is either turned on or, if the 12 volt power supply is plugged in, and the switch is in the off position, the batteries are charging. The scans shown below show that the noise source is suitable for aligning radios up to 100MHz with an ideal flat response up to 30MHz.






 This picture shows the noise source being turned on. On a monitor receiver you hear a loud rushing sound with the S-Meter registering S9 +20dB. This level remains at this level from about 200KHz to 30MHz.

 Later I'll show the results of further experiments.

The only problem in final construction I had was one of the new battery holders had an open circuit between its negative output pin and the spring contact. It looked OK but had a faulty crimp so I had to solder a link from the spring to the output tag. I shouldn't have selected these parts by lowest price!

Below is the final circuit. A1/A2 load resistors are 100 ohm surface-mount chips, 100nF are chip capacitors. I used a small 330uF electrolytic capacitor as a convenient anchor point and this is certainly not a critical value. The charging resistor isn't critical either and can be chosen to match the power supply and desired charging current. Switch S1 happened to be available and conveniently allows an OFF position for charging. Because the A1 and A2 have a maximum supply voltage of 6 volts I've added Z2 which can conveniently be a second MMSZ5V1T1G. This would really only come into play during testing when an external power supply might be used.


 I tried the noise source on my Kenwood receiver and it seems fine, but what does one see using an SDR? Basically my SDR Play receiver works just like a very sensitive spectrum analyser. It can be set to receive any 10MHz (or smaller) band between say a centre frequency of 5MHz and higher, even up to 2GHz. For the 25KHz or 100KHz scan I selected lesser bandwidths.

Below are the results. I left the vertical scale untouched but altered the bandwidth settings for the lower frequencies. Spurious signals in the noise OFF settings are due to a variety of reasons, but chiefly breakthrough from the USB cable connected to the computer together with the odd internally generated signal or indeed software artefacts (eg that centre spike).



 Above:Above: Centre frequency 25KHz, baseline -140dBm. Below: Noise -105dBm



 :Above: Centre frequency 100KHz, baseline -150dBm. Below: Noise -95dBm


 :Above: Centre frequency 100KHz, baseline -150dBm. Below: Noise -95dBm




 Above: Centre frequency 5MHz, baseline -150dBm. Below: Noise -94dBm




 Above: Centre frequency 100MHz, baseline -150dBm. Below: Noise -88dBm




 Above: Centre frequency 1GHz, baseline -150dBm. Below: Noise -99.7dBm




 Above: Centre frequency 2GHz, baseline -150dBm. Below: Noise -87.8dBm

 I must admit to be slightly confused by the various test results but suffice it to say the noise source appears to work just fine. I decided to check the total power output by connecting it to my HP431C


 According to the meter reading the power from the noise source is 1.15mW, which is pretty well 0dBm into 50 ohms or 0.25 volts RMS. This power meter has a bandwidth of 10MHz to 10GHz. The output power of the noise source I can see on my SDR is at least 25KHz to 2GHz.


Over the frequencies tested I was seeing an average of about -40dBm or 0.1uW of noise at any frequency with an RBW of 1MHz.


To bring down the power to this level that 1.15mW of noise must be spread over the full 10GHz? If so, that would result in the 0.1uW/MHz which is what I'm seeing on the spectrum analyser display?

 Below is a picture of a scan from 0-2MHz which covers most IFs. By turning on the noise source during a sweep you can see that it's producing a nice level output at a decent signal strength of about S9 +36dB.


 Now a practical test. I made a VLF low pass filter some time ago so connected this in front of an SDR. The filter has a switch to place it in or out of circuit.

 Above a scan from zero to 200KHz without the noise source switched on, baseline -150dBm, and below with the noise source turned on, top -70dBm.


 Below, with the low pass filter switched in. Not ideal as the LF end droops considerably, but the cut-off to reduce LW and MW broadcast signals isn't bad at -125dBm, and providing attenuation of (125-75)=50dB. Compare with a trace from previous tests.




 Here's a comparison between the use of the noise source with an SDR compared with the Rigol spectrum analyser and its tracking generator. Because of the low frequencies used the noise source is not ideal because it uses three 100nF capacitors in the coupling between stages. These each have an impedance of 40 ohms at 40KHz, increasing to 177 ohms at 9KHz and 800 ohms at 2KHz, hence the drop-off at the left end of the noise test compared with the high precision of the Rigol which quotes only -3dB drop at 9KHz. As there are three 100nF capacitors in series in the noise source circuit you can very roughly estimate the theoretical loss at 9KHz to be 18dB, and 30dB at 2KHz.

The Rigol tracking generator has a max output of -10dBm which is a lot higher than is usually needed but useful at times. This noise source has an output of -40dBm, but as the supply voltage is sitting at about 3 volts there's scope for increasing the output by bumping up the supply voltage to 5 volts at the expense of reduced battery life.

 See progress on the Mk2 version...


 Return to Reception