FreeDV Robustness Part 5 – FreeDV 700

We’ve just released FreeDV v0.98 GUI software, which includes the new FreeDV 700 mode. This new mode has poorer speech quality than FreeDV 1600 but is far more robust, close to SSB on low SNR fading HF channels. Mel Whitten and the test team have made contacts over 1000 km using just 1 Watt!

You can download the Windows version of FreeDV 0.98 here.

To build it you need the latest codec2-dev and fdmdv2-dev from SVN, follow the Quickstart 1 instructions in fdmdv-dev/README.txt. I’ve been cross compiling for Windows on my Ubuntu Linux machine which is a time saver for me. Thanks Richard Shaw for your help with the cmake build system.

Mel and the team have been testing the software for the past few weeks and we’ve removed most of the small UI bugs. Thanks guys! I’m working on some further improvements to the robustness which I will release in a few weeks. Once we are happy with the FreeDV 700 mode, it will be ported to the SM1000. If you have time, and gcc/embedded experience I’d love to have some help with this!

It’s sounds pretty bad at 700 bit/s but so does SSB at 0dB SNR. The new modem uses a pilot symbol assisted coherent PSK modem (FreeDV 1600 uses a differential PSK modem). The new modem also has diversity; the 7 x 75 symb/s QPSK carriers are copied to form a total of 14 half power carriers. Overall this gives us significantly lower operating point SNR than FreeDV 1600 for fading channels. However the bandwidth is a little wider (800 – 2400 Hz), lets see how that goes through real radios.

Simulations indicate it has readability 4/5 at 0dB SNR on CCIR poor (fast) fading channels. It also has a PAPR of 7dB so if your PA can handle it you can hammer out 5dB more power than FreeDV 1600 (be careful).

For those of you who are integrating FreeDV into your own applications the FreeDV API now contains the 700 bit/s mode and freedv_tx and freedv_rx have been updated to demo it. The API interface has changed, we now have variables for the number of modem and speech samples which change with the mode. The coherent PSK modem has the very strange sample rate of 7500 Hz which at this stage the user (that’s you) has to deal with (libresample is your friend).

The 700 bit/s codec (actually 650 bit/s plus 2 data bits/frame) band limits the input speech between 600 and 2200 Hz to reduce the amount of information we need to encode. This might be something we can tweak, however Mel and the team have shown we can communicate OK using this mode. Here are some samples at 1300 (the codec rate used in FreeDV 1600) and 700 bit/s with no errors for comparison.

Lots more to talk about. I’ll blog some more when I pause and take a breath.

2 thoughts on “FreeDV Robustness Part 5 – FreeDV 700”


    I had the opportunity to test the new FreeDV 700 on 40m last night under marginal band conditions. Justin, K5WAZ, and I decided to take advantage of the conditions to see if the new FreeDV 700 mode met the intended design goals for noisy/fading path communications. The test was conducted right at the end of local 40m NVIS propagation (the banded faded for NVIS approximately 10 minutes after the completion of our test). The two test stations both utilized SDR transceivers and 575′ wire loop antennas at 20 feet above the ground. Audio paths were adjusted on both ends for 0 ALC transmission. The path between the two stations is 15 miles direct, but there are numerous large mountains in the path, so the propagation was definitely via NVIS skywave. Lightning crashes were present on the band from distant thunderstorms, but they didn’t seem to affect the digital decode in any appreciable manner. The band signals fluctuated significantly as the band began to go long, so the conditions were excellent for weak/fading band conditions testing.

    We began the test utilizing FreeDV 700 mode at 50W TX power out, and copy was nearly 100%. We then reduced power to 25-30W TXPO, at which point the signal began breaking up to the point the comms were no longer able to be decoded at the RX end. From there we set 30W as the reference power for the FreeDV tests and came up with the following results:

    1) Utilizing FreeDV 700 at 30W, we saw ~3-4 dB on the FreeDV application SNR meter. Below this point, the digital signal would print on the waterfall; however, the decode did not function and the transmission was garbled. The audio quality during this portion of the test was very robotic, and at times difficult to understand due to digital artifacts in the speech; however, the transmitted message was received and understood.

    2) With that data point, we switched over to FreeDV 1600 utilizing the same TXPO. In order to achieve the same TXPO on 1600, I had to bring the transmit audio gain down 6-8dB from the 700 setting to get 0 ALC transmit. FreeDV 1600 under the test conditions at 30W was completely unusable. The signal would print faintly on the waterfall, but would not decode or even trip the FreeDV application SNR meter.

    3) We switched from FreeDV 1600 to analog SSB for a final comparison. Using flat equalization, a moderate compander setting, and 0 ALC transmit, we found that we needed ~60W average power on SSB to achieve the same general speech intelligibility as the FreeDV 700 signal at 30W.

    While the test was subjective, we concluded that the relative results obtained by comparison testing under consistent band conditions indicated that FreeDV 700 has likely met the initial design goal of offering intelligible communications over weak/fading paths where SSB and FreeDV 1600 could not, given equivalent TXPO. That doesn’t mean the 700 decoded voice was easy or pleasant to listen to; however, robotic voice with some significant digital artifacts is much better than no voice at all when the chips are truly down. We’ll continue comparison testing of this mode and report further findings as they emerge.


  2. Just had a great QSO with Gerhard OE3GBB a little while ago from here in Houston, Tx. using the new .98 version.

    At 1600 I could tell he was there but not even a syllable could make it through successfully.

    We switched to 700 and had a 15 minute solid QSO with his SNR at 0 the majority of the time.

    So a great example of where the 700 mode really makes a major difference.

    Many thanks David for this new addition as it’s working exceptionally well.

    all the best, Walter/K5WH

Comments are closed.