LinHT

Linux Handheld Transceiver

— Bruce Perens K6BP, 2025-08-12

Context:

… an amazing hardware project called LinHT, which you can see here. What is LinHT? It’s a handheld transceiver with a 500 KHz I/Q modulated SDR based on Semtech SX1255 RF-to-digital transceiver IC, capable of all modes and as much as 2 megabit per-second communications, and including a Linux system-on-module. Because production of good HT cases isn’t something we have down yet, it fits in a Retevis C62 case. [We could use a good project with industrial designers who just make HT cases, visually desirable ones, and injection molded, please, not jaggy 3D prints.] Developers can use GNU Radio flowgraphs or C/C++ on the built-in development platform in the HT! Prototypes work today and an alpha production run is being sponsored by PCBWay. If you want to start working on the platform before boards are available, the SXceiver is a good way to learn SX1255 and develop for it. There is a SoapySDR driver for it, so you can get it running with existing SDR software, or GNU Radio, immediately.

LinHT is the most important hardware project in Amateur Radio today.

Among other things, it is the perfect platform to run a variant of RADE on VHF/UHF. RADE may just be the future of voice, all voice, on Amateur Radio. It integrates what we used to use a CODEC and MODEM for into one unified component powered by machine learning (the non-hype version of AI). In tests, it works at a significantly lower signal-to-noise ratio than analog voice modes, even SSB, and legacy digital modes like D*STAR and DMR, while providing better fidelity, noise immunity, and fade resistance.

Even if RADE isn’t the next big thing, LinHT should get us there, because it provides a hardware platform capable of running new, experimental, and modified modes, and isn’t limited to just FM and FSK. Your smartphone uses complex multi-carrier modulation, HF digital modes use it, your HT should too.

I was involved in an effort to create a similar, though less capable thing, called Whitebox, with the technology of 12 years ago. The Whitebox effort failed because the design had too much electrical noise, and we spent too much time building the computer and too little working on RF. LinHT uses a pre-made single-board-computer with an ARM CPU containing two ARM A55 processors, a Mali GPU, and an Ethos-U65 microNPU to accelerate machine-learning applications, all combined on to a “stamp” form module, which is soldered onto the main PCB, and costs as little as $32 in quantity. Thus, very little of LinHT effort is invested in the computer side of the design, and the design can easily change as better single-board modules become available. And Wojciech [Kaczmarski SP5WWP] and his crew are probably better RF engineers than we were 12 years ago.