This section is part of DSP-10 User's Manual, Chapter 2 V380 - Weak Signal Operation
The PUA43 mode is designed for extremely weak communication, using very low data rates. This modes takes advantage of narrow band filtering along with long-term integration and introduces a concept called message estimation. The message lengths are a fixed number of symbols but the transmission time for the message is variable from 1 minute to as long as one has patience. In theory there is no limit to the minimum signal level. In practice the improvement over human ear/CW seems to be 20 to 25 dB, with potentially more gain for the very patient!
The following is a quick summary of the code characteristics. This is followed by information on the controls available and then the operation involved in using the PUA43 code.
The 43 symbols defined for the PUA43 code are
The letters A-Z (or a-z converted to A-Z)
The numbers 0-9
Space
Period
Comma (available for redefinition)
Forward Slash '/'
Pound Sign '#' defined as "Message Received" (redefinable)
Question Mark '?'
Dollar sign '$' to shift the meaning of the following character.
Each message can be selected as either 14 or 28 symbols. If the shorter length can be used, a 1.5 dB S/N advantage is gained. Starting from the top of each minute, the symbols are transmitted in message order, one every 2 seconds. If the message is only 14 symbols, it is repeated starting at 28 seconds into the minute. The 4 seconds at the end of the minute is reserved for a CW identifier that is automatically inserted.
The frequencies of transmission are arranged according to the locations of the spectral (FFT) bin centers. Every fourth bin frequency is used for a tone. The bins adjacent to the 43 tone frequencies are guard bins to allow for spectral broadening of the windowing functions and for any small frequency error. The bin half way between two tone frequencies is for noise estimation. Each tone estimates the spectrally-local noise by using the power in the two second-adjacent bins. The total band of FSK offset frequencies for the 43 tones depends on the SpecAnl width as follows:
SpecAnl = 1200 Hz width FSK Band = 450 to 843.7 Hz
SpecAnl = 2400 Hz width FSK Band = 450 to 1237.5 Hz
SpecAnl = 4800 Hz width FSK Band = 450 to 2025 Hz
The CW ID is always sent at FSK =375 Hz.
In order to minimize the interference from local birdie signals, a randomization is applied to the frequency of each symbol. This is coordinated between the transmitting and receiving stations by means of a random number file. Each minute the randomization "stir" is changed that shifts the tone frequencies corresponding to a particular symbol.
Provision is made for using this mode with EME work, for which it is quite suited. Doppler corrections can be automatically applied on receive and a delay of about 2.6 seconds is added to the receiver timing.
Message Estimation is used to control the display. Basically each of the 14 or 28 symbol positions has two characters printed, The large-type symbols are the "most likely" and the smaller printing is the "next most likely" set of symbols. Three colors are used to represent the likelihood that the symbol is valid. The estimated message is updated every two seconds as a new symbol is received.
Time and frequency accuracy are both important for this mode. Read the "Clock Set" section to see how the software clock for the DSP-10 can be set to within 20 or 30 milliseconds of UT. Frequency accuracy should be within 0.5, 1 or 2 Hz depending on the SpecAnl width. This requires an external reference for the DSP-10 of high quality and this same reference needs to be applied to any transverters used with the radio.
The up/down cursors move between items which become highlighted.
Radio buttons or check-boxes are changed by highlighting the item and hitting the space bar.
Text entry is at the position of the cursor in the text box.
Backspace over existing text and type in new items.
Close the dialog box by hitting Enter.
Item-by-item in the ALT-B Box:
Quality Ratio - This quantity
decides the color
division of the message estimation. It is a decision threshold
applied to the S/N power ratio of each symbol. This value divides
the low confidence color (black in the default colors) from the medium
confidence (beige). This quantity squared is the threshold between the
medium and high confidence (white) colors. A nominal value of 1.2
works well. The value can generally be 0.0 to 3.0. If the value is low,
the bright colors are easily obtained in printing the PUA43
message. A value of 3 would make the bright white letters very
difficult to obtain.
Use Random Stir- A check box to select whether random frequency shifts are used. A method to avoid "birdies" that would otherwise appear as signals
Noise Decay - Long noise averages are used to estimate S/N of each of the 43 symbols. These noise estimates are decayed to discard very old data. The Noise Decay is a number very close to 1. For message times of a few minutes a value of 0.99 is reasonable. For message times in the hours a value of 0.9999 or 0.99999 can be used.
14/28 Characters/Msg - This is the number of symbols that will be transmitted or received.
Ident String - Up to 9 characters can be entered in this text box. Each minute this string is sent in CW to identify the station. Either capital or lower-case letters will work along with any member of the CW character set.
Transmit Lines - 14 symbols are allowed on each line. The second line is not used if only 14 symbols are being sent. If any character is in a Transmit Line that is not one of the 43, a '?' will be substituted, but no warning is generated.
In PUA43 the audio portions of the receiver are fully operational, but have no interaction with the PUA43 decoding. Controls such as Filter, LMS and AF Gain can be changed at will. The same is true of the settings that affect the upper and lower spectral displays, such as Contrast, Brightness, Trace Normalization, AutoDisplay, db/div and dB Offset. The value of the SpecAve, which is the number of power spectrums averaged into each display update, is set automatically in PUA43 to the value corresponding to 2 seconds. The value of SpecAve still changes with SpecAnl width since the amount of time for each FFT changes with this setting.
Therefore, the settings that need to be made are:
Transmit base frequency - this is displayed in the Transmit Freq Box and needs to be the same at both the transmitting and receiving end of a contact.
Xmit Pwr - this mode is virtually continuous duty-cycle and the transmit power should be kept consistent with the cooling capabilities of any following amplifiers.
SpecAnl - The spectrum analysis settings are variable for Width and Windowing function:
The width of the spectral display is set with Alt-J. All FFT's are 1024 point, but the sample rate is 2400, 4800 or 9600 Hz. This produces bin resolution bandwidths, without windowing functions, of about 2.3, 4.6 and 9.2 Hz. The display in the left-hand column is "SpecAnl Window Width" where Window is the windowing functions and Width is the 1200, 2400 or 9600 Hz. Experience to date shows that the 1200 Width can be used at frequencies up to 1296 MHz. It gives better sensitivity because of the reduced noise bandwidth. As an aside, when one cuts the bin bandwidth in half, the noise bandwidth is half and the S/N of a single FFT measurement improves by 3 dB. But, each FFT measurement takes twice as long and so in any given time there is only half as much non-coherent integration. This decreases the advantage of the half-size bin bandwidth to 1.5 dB.
The windowing function (Alt-W, Alt-w or Scrl-W with kbd_alt2=1 in .CFG file) is programmable for None, Tukey-25 dB, Hamming or Blackman-harris 92 dB. These windowing functions allow trading off the selectivity of the FFT relative to the off-frequency rejection. No windowing function gives the best sensitivity and can often be used. If there are problems with strong signals or birdies, one can use a windowing function such as Tukey-25 dB, or possibly Hamming.
SpecAve - In the middle of this line is a number of either 8 or 16. This is the number of bits in the spectral data word being sent from the DSP to the PC.This MUST be 8 for PUA43 to operate (use Scrl-F2).
Clock Set - The software clock must be set for PUA43.
EME Corrections - This is toggled on and off by the Alt-L or Alt-l key and enables the Doppler correction and the 2.6 receive timing offset. The path must be selected on the bottom line as "MOON:Local Stn, Other Stn..." by use of the Srcl-F3, '<', and '' commands. This requires that the latitude and longitude for both stations has been entered into the .CFG file,
PUA43 requires the RF portions of the DSP-10. In addition, an external reference of high quality (0.5 Hz or better at the operating frequency. Current testing of this mode is being done with high quality crystal oscillators interfaced to the Brooks Shera GPS controller (July 1998 QST).
The processor load for the PC is quite a bit more intensive than for the basic modes such as USB or CW. A 486 computer operating at 50 MHz will generally support this mode, or at least that has been the experience so far.
The transmit function is continuous and the cooling needs to support this. Use the Xmit Pwr function (Scrl-O, P) to reduce the transmitter output if cooling is a problem.
Both the transmit and receive ends of a PUA43
contact must be
coordinated in
* Base transmit Frequency
* SpecAnl Width
* 14 or 28 Characters per Message
* EME or terrestrial path (Alt-L, l).
*Time within 30 milliseconds of UT.
If the path is EME, be sure that the Doppler correction is correct. The best place to check this is in the EME-2 mode. At 2-meters it is probably good enough, but as the frequency goes up the error gets greater.
The transmitting station hits the Home key to initiate transmit. At roughly the same time, the receiving station hits Ctrl-W to clear all the long-term data. Both stations should have their Alt-A boxes open that shows the status of the communication. The process is allowed to continue without any operator intervention required.
The transmit end will continue to repeat the message in FSK until the Home key toggles transmit back to receive.
At the receive end the estimated message will be displayed in the top text area. The most-likely message is displayed in large print with a cursor showing where the processing is currently taking place. Above each large letter, in small print is a second most-likely set of symbols. The confidence level is reflected in the color of the printing, with black, beige and white being low, medium and high confidence.
With time, and an adequate signal level, the
correct message will
be displayed in white print and the line above it will be random
symbols in black print. For weak signals this can take quite a
bit of time, running into the hours.
The transmit message is entered into the Alt B dialog box. Select either 14 or 28 characters and enter either 1 or 2 lines of text. You can only enter 14 characters per line. Caution: I don't have a cursor yet for the text entry lines, and there will be blanks at the end of the text line if it is not full length. If so, BACKSPACE over them and you may need to retype a character to find where you are.
Home key causes this message to be sent over-and-over, forever , unless beacon operation is in place The home key (and some others) will respond after a character is sent, just like in LHL7.
Normal operation would be to enter a new message sometime while receiving a message, but a warning: Nothing is being processed while the modal (Alt B) dialog boxes are open. So, enter your message and close the box right away. This does not apply, of course, to the non-modal Alt A info-boxes.
The two lines at the top are only the received data in PUA43. The big characters at the bottom are the "most likely" and the small one above each big one is the "second most likely." This is the "message estimation" idea. In addition, the characters change brightness, based on the likelihood of them being correct. Dark is the least likely, beige is better and white is the best.
Receive processing goes on forever and gets better as it goes. The first few minutes will show white data where there is really noise, but this will settle down.
Two parameters affect the receiver processing: 1-Quality Ratio affects the decisions about the dark/beige/white display. The range is probably 0.25 to 2.0, depending on how weak a signal you are dealing with. I start out with 1.2. Here is how this works. After a each 2 second character period, all 43 frequencies are looked at for the strongest and next strongest. In addition, the average strength and standard deviation (sigma) of the strengths is figured. If the strength is less than average + Color_scale*sigma it is printed dark. If between that level and average + 2*Quality Ratio*sigma it is printed in beige. If stronger, yet it will be white. You can change Quality Ratio at any time and it will change the display colors, i.e., it does not need to be in affect when the data was averaged.
2-Noise Decay allows for the fact that the noise level changes with time, either because the gain of the radio changes, or because the external noise changes. For each of the 43 frequencies, the noise is calculated, based on the two second adjacent FFT channels. These noise estimates are put through the equivalent of an RC decay network, with a time constant set by Noise Decay. A value of .99 is somewhere around 100 second decay time and 0.999 is around 1000 seconds, and so forth. If it doesn't average long enough, say 0.5, the noise estimate will be poor and the resulting S+N/N estimates will be poor also. If it averages too long, say 0.9999999, it will ignore the changes that are inevitable in the noise level, and end up putting too much emphasis on some data. I am using 0.99 now. Note Don't use any value greater than 1.0.
Every minute, a new random "stir" can be applied to the frequencies. This can be set in the modal dialog box ALT-B. This is displayed in the "A box",and if the transmitter and receiver are not on the same minute, nothing will be received. The stir shifts the frequency channels down by some random number between 0 and 42. If they go below 450 Hz tones, they wrap around to the highest frequency. The random number comes from a 1440 long list in the file "uhfa_43a.rnd" which is like a code key.
There is a delete character box for PUA43. It is Scrl-F10. Use the cursor keys to select the character(s) and delete/undelete with a SPACE bar. Close the box to activate the change. The readings for S+N/Ave will be strange for that character, but (I think) the character display is OK. Nothing happens to a character until its regular update time comes, ie.e, the top display is not updated after the box is closed, but the S+N energy data has been cleared.
Be sure that PUA43 is enabled by mode_mask, in
the .CFG file. It's
value is 32 and I use a mode_mask of 123 that gives everything except
LSB.
Questions and Answers about PUA43
Q - The
TICK TICK TICK drives me nuts , but
if I turn off the PC sound I don't hear the error beeps.
A- To prevent water torture, the PUA43 2-second tick can be turned on and off. This is by the .CFG variable pua43_tick (0 = no tick, 1 = ticks)
Q - What do you do if nothing is being received?
A -Check the timing against WWV. In transmit for all paths this can be done by watching the "T:tt" time indicator in the lower-left corner of the screen. This is indicating the current seconds. It should transition at the same time as the tick on WWV. In receive, this can be observed as well for terrestrial paths. The lower-left corner shows R:tt and should be exactly the same as the T:tt value. For EME paths, the receiving station needs to compare with WWV by bringing up the clock-set routine (Alt-K). The correct minute can be confirmed by comparing the Alt-A Box values for Random Stir. If these don't agree there is an error in the time or date at one end or the other. If timing is correct, check the frequencies and the other parameters.
Q -The right characters are received, but they are in the wrong position in the message. What is wrong?
A -This is the sign of a timing error. If the stations disagree by one second there can even be a duplication of some letters in adjacent positions.
Q - What is wrong when a message seems to be coming through but the characters are wrong, like W7ABC is X8BCD?
A -This is the sign of a frequency error. If the path is EME and the frequency band is above 2-meters, the problem may be a Doppler calculation error. The best way to find the RIT value, if needed is for the receive station to go back to EME-2 mode, set the bottom line for self-echoes, and find the RIT for their own echoes. Then apply this same RIT value at the receive station. Experience to date suggest that this is a good procedure to follow at the beginning of a UHF/Microwave EME contact using PUA43.
Q -Why do I get OK returns on EME-2 but nothing in PUA43?
A -Assuming that all the parameters are set correctly, it is important to note that PUA43 requires more than 5 x log10(14)=5.7 dB (for 14 symbol message) extra signal. This translates to an increase in time of at least 14 times. This is the price for only working on 14 different symbols! In addition, there may be additional increases in time due to the statistics of getting good confidence on the entire message.
Q -Copy was good on a terrestrial path and the message went to poor copy. What happened?
A -This sounds like a Boeing 747! Airplane reflections can be much stronger than the tropospheric-scatter signal. Often the Doppler shifted signal will disrupt copy. It may be necessary to clear the long-term buffers (Ctrl-W) and start again.
Q - What is the effect of microwave rain scatter on PUA43 copy?
A -It seems to depend a lot on frequency of operation and the nature of the rain. Below 1296 MHz, this is probably not a factor. At 10 GHz, rain can enhance the signal. But the Doppler shift is sometimes more to one side of the center frequency and can disrupt the copy. If the Doppler spread is close to uniform about the center tone frequencies, copy will generally be acceptable. In general, wider SpecAnl widths are better with rain.