One Billion Miles per Watt on 20m

Back in 2011 Dave, WA5DJJ, and I became interested in just how low we could go in power and still be able to copy his call letters, which is our definition of QSL for QRSS work.  On 30m we were able to make it down to 8.51 uW using image stacking technique.

Since that time we have kicked around the possibility of going down to 1 uW and last November started a new campaign to do so.

By early May of 2014 we had reached 15 uW on 30m after months of trying but could go no lower.  We need long periods of time free from interference in order to obtain a sufficient number of 10 minute grabs to use image stacking to extract the signal from the noise. Activity has increased since 2011 both from QRSS and other digital modes.  The notorious OTHR from Akritori, Cyprus is a regular visitor also.     In addition, this was a most unusual Winter here in North Florida with frequent thunder storms related to the infamous Polar Vortex which plagued the East Coast this year.

At this point we decided to try 20m to escape the QRM and hopefully see less QRN.  The difference was amazing.  Not only did we have the spectrum all to ourselves but the QRN dropped almost to nothing.  We started at 100 uW and came down rapidly in 3 dB steps to 5.8 uW.  After that the difficulty with each 3 dB step increased exponentially with the final step from 2.5 uW to 1 uW requiring 12 days until all the variables lined up in our favor.  Here's a 3-day stitched image showing our approach to the 5.8 uW level:

Note that the WA5DJJ signal was still drifting a bit until Dave installed the MEPT in a thick-walled styrofoam box.

On the night of May 17 the 1 uW signal was visible in and out on the 8 hour grabber from 0430z to 1040z.  Thirty-eight 10 minute grabs over this time period were processed with the stacking software Rot'n'Stack to produce a barely discernible image in which I could read all the letters of Dave's call.  That's 1 uW over a distance of 1164 miles/1873 km.  If you do the math that's over one billion miles per Watt for a signal propagated via the ionosphere.

Here is the results of the image stacking:

Note the use of a second signal from WA5DJJ from an older MEPT feeding a vertical.  It was not stable in time thus preventing stacking.  We used this to give some idea of propagation.  KB5R also joined in for the same purpose.  The cw signal from IK6ZEW was strong but not stackable.  The 1 uW signal eminated from a QRP Labs U2 mept feeding a Cushcraft A3 triband Yagi.  The receiving antenna at W4HBK was a 30/40m inverted V with the apex at 18 meters.

For comparison here is one of the better 10 minute grabs:

It's fun to compare our miles/Watt to that of the NASA's deep space probe Voyager 1 which at a distance of 11.8 billion miles from Earth running a 20 Watt transmitter which gives 590 million miles per Watt.  Of course they are sending HiDef images and telemetry compared to our 6 scratchy letters but on the other hand they are using a 70m dish at the receiving site.

Here is the story from Dave's point of view:

http://www.zianet.com/dhassall/BILLIONMPW.html

de w4hbk

Frequency Scale Calibration for QRSS Grabbers

Spectrum Lab provides two methods of frequency calibration which can be utilized to set the frequency scale accurately on all bands.  The first applies to calibration of the sound card by injecting a known audio frequency and comparing it to the frequency measured by SL.  Figure 1 shows the Audio I/O screen for entering these values in the "Correct Frequency" and "Displayed Frequency" boxes on the right.  After entering click on the "Calibrate Input SR" and the sound card's Sample Rate will be determined and entered in the "Sound Rate Calibration Table".  I generally use a nominal SR of 11025 Hz and in the actual value is 11099.8922552 Hz which is now applied to the SL measurement.  At this point Spectrum Lab is reading audio frequencies accurately.  I used the standard tones of 500 and 600 Hz provided on the WWV carrier when listening in AM mode.

Figure 1.  Spectrum Lab Audio I/O Screen

The second frequency adjustment is via the "Radio Frequency Offset" value found on the Spectrum(2) screen, Figure 2.  This is the frequency to which the radio is tuned and adjusts the scale to read in units of the actual frequency being received.  For example, on 30m, my rx dial is set to 10138.70 kHz to produce an audio tone of 1300 Hz for 10140 kHz.  If the master oscillator were perfect then the frequencies measured by SL would also be perfect.  However, even the best receivers are off by at least a few Hz (unless locked to GPS or an atomic standard) so a little extra correction is needed.  This can be applied by adding or subtracting the appropriate value to the rx display.  My rx on 30m reads low by 7 Hz and I add this to the offset value as 10138.707.

Figure 2.  Spectrum Lab Radio Frequency Offset 

The offset can be determined in general by measuring the error for several know frequencies and applying a linear regression.  This is possible because the frequency my TS-440 and as far as I know all modern transceivers is controlled by a single crystal.  I used the standard frequencies of 5, 10, 15 and 20 MHz from WWV in Colorado to generate the curve shown in Figure 3.  The linear regression for this data is:

            F(offset), Hz = -0.2 + 0.672*F(rx), MHz,

where F(offset) is the offset needed to correct the received frequency and F(rx) is the receiver frequency.  Ideally the intercept value should be zero but the value of -0.2 is easily attributable to measurement error.  The offset values for each band are also printed on Figure 3 for quick reference.

Figure 3.  Offset Graph and Table

Use of WWV for frequency calibration can be tricky due to Doppler effects caused by motion of the ionosphere.  I made my measurements on a day when ionospheric conditions were settled and at a time when the night-to-day movement should have been completed.  I confirmed this by monitoring the Doppler shift for several hours.

Comparison of SID detectors

On 07NOV13 there was a Sudden Ionospheric Disturbance which was captured by three amateur receiving stations:

G4JVF VLF receiving NAA on 24 kHz
W4HBK grabber on 10m receiving G0PKT
W5COV receiving WWVB

This is a comparison of the exact timing of the onset of the SID as seen by the three observers.

G4JVF VLF RX Receiving NAA in Maine

W4HBK 10m Grabber Receiving G0PKT

W5COV VLF RX receiving WWVB in Colorado

I measured the times on the first two by measuring distances with a ruler and interpolating.  Marks estimating the onset were placed first and then the measurements taken.  On the last image since the SID wasn't clear to me I calculated the time estimate from the first two and placed that on the figure.  As I understand it's one of the small spikes.

For comparison here is the X-Ray flux measured by the GOES satellite:

GOES X-Ray Flux for 06NOV13 SID

All things considered the agreement is excellent.  At the time of occurrence the Sun was over the Atlantic which would strongly illuminate the skip points for the EU-US path.   The W5COV-WWVB path was almost in the dark and might be expected to be minimally effected by the SID.

Previously with SIDs I had gotten the impression that the QRSS method was more sensitive to the onset time than were other methods.  But now that I have seen them together and put some numbers with onset features  it looks like there is no significant difference.

de w4hbk

An Automatic A/B Antenna Switch

A while back I did some tests with VK1OD to compare antennas using an automatic antenna switch he had devised for antenna testing.

I made a version of Owen's antenna switch which can be controlled by the QRS keying program  which I use to key my mept via the computer's serial line.  In the experiment described in this post it is being used to switch in an attenuator so I can transmit different power levels on the Mark and Space of the QRSS FSK signal.

Here's the schematic,

The driver transistor is located in my mept box and connected to the A/B switch box by a twisted pair.  DTR is the DTR line of the serial cable coming from the computer.  My mept is also keyed from this connection via another driver transistor

Here's a picture of the resulting keyer:

The attenuator is an L design with 50 Ohms to ground and a series resistor of 100 Ohms between A and B to give an attenuation of 10 dB.  You can see where I tack-soldered in parallel resistors to reach 10 dB as measured using Spectrum Lab.  The twisted pair on the left is the keying line from the serial port and there is an LED to indicate when antenna port B is switched in.

When used with the QRS keying program the switch automatically reduces the 1 W signal to 100 mW so one is on the Mark and the other on the Space.

The other use for this switch will be to compare antennas, my next project.

I have found the 10 dB comparison most illuminating.  I seldom run my mept but when I do I generally use   1 W because I'm interested in observing propagation and want to ensure results.  But, I often wonder what my signal would have looked like at the typical QRSS power level of 100 mW.  Now I know.  More often than not the 100 mW level is in the noise compared to the 1 W and would have required image stacking to identify the call.  It's not just a matter of power because of other factors such as the grabber's antenna and local noise and how the waterfall display is set up.

Here's an example taken from PA2OHH's grabber.  In the first image which is a single grab the 100 mW Space is barely discernible and not clear enough to ID the call which is clearly seen in the Mark:

Here's a stacked image of multiple grabs from the same grabber:

Now you can see the 100 mW Space quite clearly and read the call in un-cw.  I use this technique often on 40m where the 'barefoot' mepts don't make it out of the noise well enough to be identified.

It's no surprise that a 10 dB power gain makes a big difference but if you can read the lower power level then the gain is not very impressive.  It's when the lower power is not visible at all and them emerges from the noise to be readable at the higher level that the difference is most impressive.

de w4hbk

A Comparison of QRSS and WSPR for Measurement of Signal Strengths

This post describes measurements I made with the help of Keith, G6NHU, to compare Signal-to-Noise measurements displayed by WSPR with those I make using QRSS using the Spectrum Lab spectrogram technique described in a previous post

The versatile Ultimate 2 from QRP Labs has the capability of transmitting both QRSS and WSPR and can do so in the same message frame.  G6NHU has been doing this with his U2 for some time now by designing a message that sends QRSS for 8 minutes followed by a 2 minute WSPR frame, making it possible to monitor both modes on a routine basis.  Here's what it looks like on 20m:

Figure 1.  G6NHU QRSS/WSPR message

Keith usually transmits WSPR on the WSPR frequencies but for the purpose of this experiment it was on the same frequency as the QRSS otherwise it could easily disappear in the crowded WSPR band.  Note also that the WSPR message has the same bandwidth as the 5 Hz FSK so doesn't cause problems to other QRSS stations.

For our tests I ran the WSPR program and recorded G6NHU while simultaneously making grabs of the QRSS part with this screen:

Figure 2.  QRSS Screen Used to Measure SNR

The QRSS measurement is based on the SL "long term average" function which in this case is the 2 minute period at the end of the QRSS message shown in Figure 2.   The WSPR measurement immediately followed the 2 minute-averaged grab.  Refer to he link give above for more details.

Over the time frame from 1900z to 0100z on 20m I recorded about 30 instances where G6NHU appeared on WSPR and had a clean grab on the Pensacola Snapper near the same time.  Not having much to do an a very rainy day I read all the QRSS captures and made a spreadsheet of QRSS/WSPR SNR determinations. From this I plotted the SNR's measured via QRSS and WSPR:

Figure 4.  Plot of SNR's Determined by QRSS and WSPR

The big difference in scale is caused by the bandwidths used in WSPR and QRSS.  WSPR relates the SNR to a BW of 2500 Hz while my Spectrum Lab settings are for a noise BW of 0.25 Hz. SNR is inversely proportional to BW because noise power varies with BW.  Therefore the change in SNR expressed in dB is,

delta(SNR)  = 10log((S1/N1)/(S2/N2)) = 10log(BW2/BW1), dB

     note that S1=S2 since the signal is monochromatic and does not change with BW

Thus the change in SNR for 2500 Hz and 0.25 Hz bandwidth's is:

         10log(2500/0.25) = 40 dB

which is added to the WSPR data in Figure 3 to obtain that inFigure 4.

Figure 4.  Correction for Bandwidth Differences

One thing in favor of the QRSS method is that it's possible to actually view the QRN/QRM situation at each determination of SNR...see Figure 2.  Compare this to the busy WSPR spectrum in Figure 1 where interference seems to be a distinct possibility.  As described above, the QRSS and WSPR measurements are 2 minutes apart which may account for some of the differences in agreement.

The large fluctuations in measured SNR are due mainly to QSB and not system noise.

de w4hbk

Spectrum Averaging to Improve Measurement of Signal Strengths

Many grabbers based on Spectrum Lab include a spectrogram along with the waterfall to give some idea of signal strength. I have recently enhanced this feature on my grabber to facilitate accurate measurement of signal strengths. Spectrum Lab makes spectrum determinations at a defined scroll interval which in the case of my 10 minute grabber is 0.7 seconds. The long term average function, selectable in the Spectrum(1) configuration window, averages during the grab interval to produce a dramatic reduction in noise. For the 10 minute grab interval this results in 857 individual spectra being averaged, resulting at a 15 dB improvement in signal-to-noise. Here's what it looks like (yellow is the long term average, red a single spectrum):

FIGURE 1

There is a problem in reading QRSS signal strengths because of the dual FSK peaks. If the signal was unmodulated, i.e., a straight line, continuous carrier then the displayed peak would be meaningful. However, an FSK signal has two peaks for Mark and Space which under the ideal condition of no QSB would be different due to the different cumulative times of each. To get the true signal level requires a little math to add the two, doing so in linear space rather than dB space. It's necessary to take the antilog then add and take the log of the results to get back to dBs. The following sketch shows how this works:

Figure 2

This is a bit of effort but not too bad once a work flow is established. It's worth it when you bear in mind that we are measuring signal strength over a real dx path with a precision approaching several dBs. Here's a grab I used to compare the signal strength of 4 stations located in England (a,b,c,d) and 2 in the US:

Figure 3

It is not wise to compare signals with large differences in path length because of unequal Doppler broadening as will be discussed below. This table shows the raw data, mark, space and noise, along with the processed results and comparisons:

Table

The noise baseline was subtracted from Ptotal to get the SNR. We could conclude from this data that on the US side, station f was 7 dB stronger than station e while on the European side, how other stations compared to station b which is consistently the strongest EU station seen on my grabber. Repeated observations of the EU stations give about the same results from day to day which would allow antenna experiments to improve one's signal.

This particular grabber screen was designed especially for this kind of measurement with a larger spectrum window for more easily reading the dB level and with a 3 minute grab interval to better follow the QSB.  

There are some problems and limitations to be considered. First, signals must be continuous, either a steady unmodulated carrier or FSK without an off time during the grab period. Second, the signals must have a meaningful, defined peak. Hellschreiber and drifting signals will not work because there is not a defined peak. Third, the width of the signals being compared must be the same otherwise we'd have to consider the area under the curve which is beyond the capability of this technique. Doppler broadening generally increases with path length so comparisons should be made only signals from the same general area. Finally, the signals must be continuous The signal at g is not usable for this reason.

As a bonus, I have adjusted the output of my receiver to make the noise floor always at -110 dB so that day-to-day or perhaps even seasonal comparisons are possible, assuming that the rx noise floor and gain/loss of the antenna/feedline remains constant and I see no reason why they shouldn't. Referring to Figure 2, in the spectrogram the reason there is so much space to the left of the noise baseline is so I can see the -110 dB level to make this adjustment, which I do while substituting a 50 Ohm load for the antenna. As long as I don't fiddle with the gain controls the noise floor is remarkably stable from day to day.

Some stations using the U2 "Ultimate" from QRP labs are transmitting both QRSS and WSPR signals by time sharing where QRSS runs for 8 minutes then shifts to WSPR for 2 minutes. I have made some preliminary comparisons to signal strengths made using the technique described here with the dB levels reported by WSPR and get good agreement. When I have made more comparisons and am confidant in the results I'll present it in another post to this blog.

In case you're wondering, my receiving system is linear because the AGC has been disabled using a technique described in a previous post.

de w4hbk

Plotting Signal Strengths with Spectrum Lab

Spectrum Lab has a feature which allows one to bracket a signal on the waterfall display and watch its signal strength as a strip chart recording.  This feature is called "Watch List & Plot Window" and is found in the "View/Windows" menu at the top of the display.  Here's the set-up screen:

Set-up screen for SpectrumLab signal plotter.

It's a bit tedious to learn to use and involves a set of commands, functions and expressions reminiscent of BASIC which allow a signal on the waterfall display to be selected and plotted.  But the good news is that only a few instructions are necessary for making a simple strip chart recorder.  I use two expressions:  noise_n(f1,f2) which plots the band noise in the range between f1 and f2 which are two audio frequencies on the WF display and peak_a(f1,f2) which yields the peak signal between two frequencies.  Up to 6 signals may be selected and plotted at the same time.

The noise_n function divides the range between the two specified frequencies into 256 bins and averages the 1/4 lowest values to obtain the band noise.  The likelihood of including actual signals in this average is very low unless there is broadband interference present such as from RTTY or JT65.

The peak_a(f1,f2) expression brackets the signal of interest, divides it into 256 bins and selects the largest value to plot.  This makes a varying QRSS fskcw signal appear as a continuous line since the mark and space are always within the brackets.  Wolf thinks of everything.

OK, so why the interest in plotting signal strength?  One reason would be to watch for changes in your signal while doing some antenna comparisons, for example.  Another and the one in which I'm interested is to study QSB.  There are three causes of QSB:  multipath interference, Faraday rotation and ionospheric focusing.  In multipath the signal is reflected from two or more layers and arrives at the rx antenna to interfere destructively and constructively.  In Faraday rotation the polarization of a linear transmitted wave rotates as it passes thru the Earth's magnetic field and arrives at the polarized rx antenna at a state other than of the tx antenna, ie a vertically polarized transmitted wave can arrive at the rx as another polarization, including eliptical.  Focusing is cause by the shape of the ionosphere which can vary from convex to concave.

Scientist and engineers who study propagation have utilized simultaneous recordings of the vertical and horizontal vectors to detect the Faraday effect and the qualitative nature of the overall variations to separate the three effects.  It ain't easy and requires sophisticated computing capability and understanding of propagation, definitely a job for professionals.  However, that doesn't keep we amateurs from having fun with the subject, making our own observations and attempting to guess which effects are occurring.

My favorite reference on QSB is the dissertation of Dr. Yau at the University of Adelaide who studied the effect on OTHR signals in the 6 to 10 MHz range.  There's a good discussion of hf propagation there plus a details of the receiving apparatus and computational techniques.  Imagine getting a PhD in QSB!

Here's an example of QSB recordings I've made on some 30m signals:

Example of strip chart recording

The Red trace is the band noise, Yellow is KD4PBJ near Chattanooga, TN and Green is AK4T near Atlanta, GA.  

de w4hbk