Next on the bench is the low-pass filter, which is mostly responsible for curtailing the harmonics generated by the VFO and mixer. Since limiting out-of-band signals and noise is also useful on receive, the LPF is connected directly to the antenna socket to be useful in both modes.
This LPF is particularly simple, using just three RF chokes and six capacitors. VK3YE’s original low pass filter used a couple components I didn’t have (namely the 820pF caps), so I played around a little in LTSpice to work out a viable filter with the components I had. Here’s what I ended up with:
SPICE simulations show that this filter maintains a similar low-pass characteristic to the original VK3YE design. Moreover, the insertion loss is approximately equal. The match to 50-ohms isn’t great, but is acceptable. And perhaps most importantly, attenuation at the second harmonic is strong at about 42dB:
Constructing the LPF was simple – just six capacitors and three molded chokes that I picked up on a recent trip to California. I laid the filter out on a small piece of scrap copper clad. I enjoy when the layout of a board can highlight the symmetry of circuit itself.
Before attaching the filter to the rest of the project modules, I did a quick manual sweep of it with my SI5351 board. It’s somehow very satisfying to roll the frequency up past the filter knee and see the higher frequencies just drop off. One word of caution – the SI5351 doesn’t have perfectly uniform output power across all frequencies, so I put one channel of the scope on the input of the filter and one on the output, to observe relative attenuation. Of course, terminating the filter is a must.
Back in March, QST magazine published an article about modding a cheap Forty-9er kit from eBay to incorporate a digital VFO. The original Forty-9er was a kit from the NorCal QRP club, a 40m transceiver designed to run on a 9V battery, hence the name. It, like many other kits, is based around an NE602 and an LM386. In the last couple years, cheap kits bearing the same name have been appearing on eBay, which bear only a scant resemblance to the original. The biggest difference is that where the original kit had a VXO attached to the NE602 mixer, the eBay kits are designed to be rock-bound to a single frequency. Perhaps this was the motivator for the QST article, to restore some frequency coverage to these fixed-frequency kits.
The process of using a digital VFO with an NE602 architecture is simple enough. Specifically, an AD9850 breakout boardis used to provide the signal, and a small BJT amplifier increases the power output from the DDS chip. The oscillator power is adjusted to show about 300mV P-P in-situ. After the article was published, one of the authors, K2ZIA produced a limited run of kit boards, which include the amplifier and sockets for both the AD9850 board and an Arduino Nano to control the DDS over i2c.
I purchased the breakout board, AD9850, Forty-9er kit, Arduino Nano, and a rotary encoder from a fellow ham, Justin AJ2Q, who had gathered the pieces but was focusing on other pursuits and wanted to pass the project on. Final assembly was pretty straightforward; there are a couple of mods that need to be made to the Forty-9er (specifically, replacing the oscillator crystal with an input for the VFO, and swapping the crystal input bandpass filter for a much-wider two-element bandpass filter), and connection made so that the Arduino can detect when the key is down and shift frequency. AJ2Q had already done most of this, so only some final tweaks and cleaning up some soldering were necessary.
The original Arduino code was designed to make use of a 16×2 LCD display to display the current frequency, as well as licensing information according to the 40m band plan. Since I’ve already been playing with an LCD display on another ongoing project, and since I wanted this to be a simple and durable bit of kit, I wrote a bit of code that instead flashes the current frequency in morse code on a panel-mounted LED. The display is triggered by the press of a button. The number of digits to display is configurable. I’ve found that just displaying the three kilohertz digits is plenty (I don’t need to be reminded I’m on the 7 MHz band every time, and I don’t need precision better than KHz for simple operating). As always, you can see the code on Github.
The full schematic of the original Forty-9er as well as the necessary mods can be found on Farruk K2ZIA’s website. The only additional hardware changes I made were to wire an LED and a 1k resistor between Arduino pin 14 and ground for the LED, and an SPST button between Arduino pin 4 and ground to trigger the frequency display. Like the original code, depressing the encoder changes the tuning rate, though I modified the possible step values to be only 1000Hz, 100Hz, and 10Hz, in that order. 1KHz is useful for zooming around the band, 100 Hz is useful for tuning a specific signal, and the 10Hz step is mostly for resolving SSB/DSB/AM signals cleanly.
On the rear of the radio are the BNC antenna jack and the power pole power input. (Useful tidbit – a pair of connected 30A powerpoles fit neatly in the cut-out for a VGA connector!) I’ve yet to fashion a front-panel, so the connections on the Forty-9er board for a key and headphones are directly accessible.
I’ve had the rig out to the park a handful of times now, and it sounds good! Like my other direct-conversion NE602-LM386 experiments, the audio quality is great, but broad as barge, so selectivity suffers. The sidetone is clean, at around 700Hz, from the Forty-9er’s little BJT oscillator. I enjoy being able to tune up into the phone portion of the band and listen to SSB QSOs and nets and such, which provide a good sketch of current propagation conditions. Being able to quickly switch between tuning steps is helpful, as I tune around and try to find someone sending CW slow enough for me to keep up.
The only major limitation of the rig is that, with only a single-resonator input bandpass filter, out-of-band signals can get into the radio and cause interference. Specifically, World Harvest Radio WHRI, who maintain and 500KW (no that’s not a typo) transmitter in South Carolina, is often audible everywhere on the dial, which is distracting at best. A stiffer bandpass filter will be necessary soon. I’ve also been experimenting with a peaked audio filter to help with reception of CW, but that’s still experimental.
The unit puts out about 3 Watts, which proved to be enough to make my first CW contact and, over the weekend, my second. The gentleman on the other end this time was Gary N4PIR, who was running 5W on his FT-817 into a trapped vertical. I think a little afterburner would be helpful on my end, as we were definitely fighting QSB. But that’s a later project.
Somewhere in the near future, I’m hoping to experiment with building my first superheterodyne receiver. While I may be putting the cart before the horse in working on new receivers before putting up a proper antenna, until the weather kindly agrees to stay nice for more than one day at a time, it’s the soldering station for me.
An essential part of a superhet receiver, and the part I’m excited to experiment with, is the IF filter. The matching (or at least characterization) of the motional parameters of a few crystals is critical for filter design, so to that end, I put together a simple test oscillator a la G3UUR to try matching some crystals.
The G3UUR is a simple Colpitts oscillator with a buffer, which you are intended to attach to a frequency counter. A small bit of machine header allows for the insertion of individual crystals for testing. A SPST switch allows for one lead of the crystal under test to either be shorted to ground, or grounded through a small capacitor. By measuring how much the frequency of the crystal is pulled by added extra capacitance to the oscillator, we can calculate some basic parameters of the filter, namely the equivalent series capacitance and inductance.
The circuit went together fairly simply using parts out of my junk box. This was my first time trying out “island pad” construction, making isolated islands on the copper clad with a drill or scribe. W2AEW has a nice example of planning island pad layout starting around 1:45 in this video, and Jack K8ZOA has a really nice write-up of the construction process on his Clifton Laboratories page.. Here you can see my sketching process as I roughed out the transition from schematic to island pad layout.
Since I don’t have a nice diamond core drill bit as suggested by others, I made my own sloppy island pads using a engraving bit and my cheapo Harbor Freight knock-off Dremel. Though it’s not a particularly high powered tool, it eats through the copper on copper-clad with ease. Would definitely use this method again, it was tremendously easy. It also allows for the creation of oddly shaped islands, sad for making loops to attach power or ground leads to.
With the board carved up, it was easy enough to solder the components in place. I turned my iron up to 850 degrees to help tack the components onto the copper board quickly, and the soldering process only took about 30 minutes.
As a frequency counter, I’m using a cheapie “cymometer” I got from ebay. While not terribly robust, it’s easy to use and seems to give fairly good accuracy.
The G3UUR method is very simple: place the crystal in the oscillator with its lead grounded, and read off the frequency on a frequency counter. We’ll call this the base frequency, f. Then open the switch, so that the crystal is only grounded through the capacitor Cs, and read off the frequency again. (It’s worth measuring the capacitance here, rather than trusting the labelled capacitance.) The difference between the two frequencies we’ll call ΔF. The capacitance of the crystal itself (measured across the leads with a meter). we’ll call C0. The motional parameters of the crystal are then:
A quick experiment on a miscellaneous 9.838 mhz XTAL I had lying around gave results of 11 fF and 23.8mH, which are within the realm of possibility for a crystal of this size. More encouragingly, a June 2006 paper from Jack Smith K8ZOA suggests that the simple G3UUR method seems to give results within about 5% of more complicated and exact methods, like using a calibrated Vector Network Analyzer or performing phase comparisons of frequencies across the crystal.
Hopefully this is the first step toward building a “good enough” three or four crystal filter for a superhet receiver.
As wonderful as it was to ‘complete’ the Virgin Receiver and have a set of ears to start listening on, more and more I find myself wishing for the ability to change frequency and explore the 40m band a bit. To that end, I’ve been experimenting with hooking my Si5351-Based VFO directly to the Virgin as an easy way of giving it the ability to QSY.
(The VFO, as it happens, has a few more features nowadays, including being able to act as WSPR beacon and having an auto-CQ mode. The code, as always, is on Github.)
Here’s the receiver as built:
Jason NT7S dropped a hint in one of his older blog posts that the Ne602 likes to see about 300 mV p-p when being driven externally, and that a 10dB pad could be used to bring the Si5351 down to this level. To that end, I put together a quick Pi-attenuator consisting of one 120-Ohm and two 150-ohm resistors.
The attenuator plugs directly into the receiver where the crystal X1 usually sits, in a little 1×3 piece of female header. The output of the 10dB pad plugs into the side of the header that’s connected directly to pin 6 on the NE602; the ground on the attenuator plugs into the other side of the header, and is therefore connector to ground on the receiver.
The longer red and white wires you can see attached to the pad connect to ground and the CLK0 output of the Si5351. Here’s the current setup, spread out on the bench.
It’s already quite successful! I can scan up and down the 40m band and pick out CW signals pretty well. SSB signals are faint and pretty un-hearable. Unfortunately, for some reason, my receiver has also turned into an AM radio:
I think some kind of high-pass filter is in order here. There’s always been a bit of AM bleedthrough with this receiver, but given that the AM stations are attenuated by the RF Pot on the receiver, it seems like most of this signal is coming in through the antenna and not, say, bleeding into the audio amp.
With an eye toward getting my rudimentary Si5351-based transmitter on the air, I’ve started putting together a number of low pass filters to knock down some of the nasty harmonics that come with using a clock chip as a frequency generator.
The filters are all based on the designs given by W3NQN in the GQRP technical pages from 2015. (This is another point at which I’m taking design inspiration from the QRP-Labs QRSS beacon.) The paper lists both the target values for each of the toroids and capacitors for a 7-pole low pass filter for each band, as well as the required number of turns for each toroid and an appropriate toroid to use. For this project, since my target frequencies range from about 7 MHz to 14 MHz, I’m using all T37-6 toroids from KitsAndParts.com. The basic design of each filter is the same:
Here’s the values that W3NQN gives for low pass filters for the 20m, 30m, and 40m bands:
Unfortunately, I don’t have exactly the specified capacitors in my kit, so here’s my approximation to these values. (Note that the number of turns is different at 40m than W3NQN recommends, to more closely match the specified inductance). All inductors are wound on a T37-6:
L2, L6 (turns)
1000+220 (series) = 180pF
1000 + 690 (series) = 408 pF
16T = .77uH
17T = 0.87uH
220+47 = 267pF
220+220+100 = 540pF
19T = 1.08uH
20T = 1.20uH
21T = 1.32
24T = 1.73uH
Some quick simulations in LT Spice confirms that these little adjustments don’t have an enormous effect on the filter’s behavior. On 40M, for example, it has the effect of moving the cutoff frequency down maybe 100 KHz:
So far, I’ve only constructed the 40m filter. It’s built on a little 2.3cm x 10cm piece of coppper clad. (I’ve got a bulk stock of 7cmx10cm pieces, which cut nicely into threes to make little projects like these).
The filter seems to be working great – a little qualitative examination with an oscillloscope shows that the filters are smoothing out the waveform at 7 Mhz quite nicely. Above that, the attention seems to match the slope predicted in the simulation above.
With this filter on my little homebrew transmitter, I finally made my first (official) WSPR contact… but that’s a story for another post.
LTSpice is a fairly approachable, very powerful circuit simulation software system. It has a straightforward GUI interface – drop in parts from catalog, draw wires, run simulation – with a strong simulation engine sitting behind it. I downloaded it this week for the first time, and it’s already proved its worth.
Many others have written excellent tutorials about using LTSpice, so I won’t rehash the details here. I found the tutorial on All About Circuits to be very useful, if a little bit hard to follow.
One of sample projects that that tutorial walks you through is analyzing the response of a basic LC bandpass filter using LTSpice’s AC Sweep mode. After working through that, I thought it might be fun to model the front end filter on my little Direct Conversion receiver to see what the actual (expected) frequency response is. Since I had ripped those values straight from the GQRP Sudden, it seemed like a good learning experience.
So, I quickly drew up a little model of that front end filter. It looks something like this:
And, running a sweep from 1 MHz to 10 Mhz, I got… a steaming pile of what the heck kind of input filter is this??
-24dB at best, a “peak” somewhere above 25 MHz, there’s no way the the GQRP would have put this out in a kit. I went back through my network, looking for a dropped connection, or a capacitor with a value in farads instead of picofarads (boy will that cut your gain). But no dice, the circuit seemed solid…
…Until the third time I checked the component values, and noticed one fatal slip-up. You see, R1 and R2 in the above schematic represent the source impedance and the input impedance, respectively. While the source impedance could be a nominal 50 ohms, the RF input impedance for the NE602 mixer is at least 1.5 kOhms. Doh!
So, with a slight change to R2 in the schematic, and re-running the simulation:
Now that’s more like it! A nice peak in the middle – not a ton of attenuation out of band, but pretty good for a little five element filter. We can even zoom in on that peak a little
It looks like my filter’s currently peaked at around 6.63 MHz, instead of the ~7.15 MHz I would have hoped for, but considering the lower and upper boundaries of the US 40m band are only 2 and 4 dB down from the peak respectively, I’d call that a good first start.
A little more messing around showed that using a proper 10pF series capacitor instead of ~9.8pF doesn’t make much difference. The best solution I found to be to reduce the shunt inductors to about 4.8 uF, which puts the filter peak at just over 7 MHz. Of course, this assumes all ideal values and no parasitic effects for any of the components. And that my hand-wound toroids actually match these inductances. But hey, it’s somewhere to start!
It turns out, LTSpice isn’t just useful as a software tool – it makes a great rubber-duck debugger as well. While squinting down at the board to double-check the value of those tiny series capacitors, I realized I’d made a horrible assembly error. I should have put those two 200pF shunt capacitors in series on either side, as in the above diagram, but I instead put them in parallel….
Yeah, forget the audio pre-amp for now. I’m losing 30+ dB to my front end filter. Time to heat up the iron.