Daniel Estévez
LightNews LightNews
banner
destevez.net
Daniel Estévez
@destevez.net
630 followers 65 following 280 posts
Everything space & RF. Amateur radio operator (EA4GPZ / M0HXM). PhD in Mathematics from Univ. Autónoma de Madrid. he/him
Posts Replies Media Videos
destevez.net
I have made a new release of the Maia SDR firmware. This is a maintenance release that updates dependencies and includes some USB Ethernet connectivity improvements to Android devices.
The release changelog. Update Maia SDR to v0.12.0. Updates dependencies. Update IQEngine to latest main. Add IPv6 link-local address fe80::1/64 to simplify connecting to the Pluto by ssh before the IPv4 network is configured (for instance this can be used in Android devices that use an address of the form 10.xxx.yyy.zzz/24, which is not covered by the ipaddrmulti feature) Include IP from fw_printenv ipaddr in the Subject Alternative Names of the maia-httpd SSL certificate.
November 9, 2025 at 7:21 PM
destevez.net
New blog post: sigmf-toolkit. I present a new Python package I've made and showcase one of its tools, which adds to a SigMF file annotations for each packet in a PCAP file. This is a good way of doing protocol analysis and debugging across the physical and upper layers. destevez.net/2025/11/sigm...
A screenshot of Inspectrum that shows a waterfall containing multiple packets. Each packet is bounded with an annotation. There are also annotations for reply packets, which are not visible in the waterfall. The mouse is hovered over one packet, and the tooltip shows a dump of an IPv4 ping packet generated with Scapy.
November 2, 2025 at 2:14 PM
destevez.net
New blog post: Non-coherent m-FSK BER. Another short post in which I derive the following formula for the symbol error rate of non-coherent m-FSK and validate it using simulations. destevez.net/2025/10/non-...
A formula of the symbol error rate for m-FSK. It is a summation with m - 1 terms that contains a binomial coefficient and an exponential term that depends on the Es/N0. A plot showing the BER versus Eb/N0 of m-FSK for m=2,4,8,16,32. The results using the closed-form formula and simulations are compared.
October 25, 2025 at 9:28 AM
destevez.net
When setup timing is violated, typically the output delay becomes much larger than the nominal one, increasingly more as the data transition approaches the clock edge. This goes somewhat against the usual idea that setup violations mainly cause metastability.

Read more: destevez.net/2025/09/flip...
A simulation of a setup timing violation. A data transition from zero to one is done increasingly close to the clock rising edge. We can see the corresponding outputs in traces of different colours. As the data transition approaches the clock edge, the output delay increases a lot, until the new data is no longer captured.
September 28, 2025 at 9:03 PM
destevez.net
Using ngspice, the timing properties of the flip-flop can be measured. Here is an output delay simulation, where the output for different process corners is shown in different colours.
A simulation of output delay. Two clock cycles are simulated, and the data input first is one and then is zero, so we can see how the input data propagates to the output after the clock rising edge. Different process corners are shown with different colours in the flip-flop output voltage trace, so we can see the difference between the fastest and slowest corners.
September 28, 2025 at 9:03 PM
destevez.net
New blog post: Flip-flop timing experiments. I design an ASIC flip-flop using Magic and play around with some ngspice timing simulations of output delay, setup and hold. I have made my flip-flop (top) as small as possible, so it is half the area as the standard cell (bottom).
Two ASIC cells. The one on the top, which is the flip-flop I have designed is half as tall as the one on the bottom, which is the equivalent cell from the standard library. My flip-flop is also slightly narrower. There is a large difference between the transistor size in both cells, and also the amount of room available to route all the signals. The standard cell can route everything in one metal layer, while my flip-flop needs three metal layers.
September 28, 2025 at 9:03 PM
destevez.net
New blog post: 10 GHz FMCW radar LO board. I've finished assembling and testing my ADF4158+HMC391 3.9-4.45 GHz LO board, for which I asked for reviews a few months ago. In the post I explain some aspects of the board design, and the software I've written and testing I've done.
A photo of an assembled PCB. It is a small board with an SMA connector, a few surface mount chips, a pin header, and tiny 0402 passives.
September 4, 2025 at 9:42 AM
destevez.net
In this post I dive into the 13 spectral bands captured by Sentinel-2 and explain how they can be used to measure vegetation health and fire affectance. This image shows SWIR in red, highlighting burnt areas, NIR in green, highlighting vegetation, and visible green in blue.
A false colour image that shows short wave infrared in red, near infrared in green and visible green in blue. Vegetation looks green or bright green, depending on density. The burnt area looks red, except for some surviving vegetation which can still be seen in green.
August 19, 2025 at 10:41 AM
destevez.net
New blog post: Multispectral analysis of the Tres Cantos wildfire with Sentinel-2 data. This image shows vegetation surviving the fire in green, vegetation lost by the fire in magenta, and areas affected by the fire but with less vegetation loss (dry grass) in blue.
A false colour image that shows vegetation surviving the fire in green, vegetation lost by the fire in magenta and burnt areas of low vegetation in blue. In green we can see the parks in Tres Cantos, and the creeks and forests highlighted. In magenta we see a large part of the Monte de Viñuelas woodland, which has been burnt severely, as well as some creeks badly affected by the fire. In blue there are areas of dry grass pastures that have burnt, mostly outside of Monte de Viñuelas and northeast of Tres Cantos. There are spotty blue-green areas in the eastern part of Monte de Viñuelas, showing partial fire affectation with trees surviving.
August 19, 2025 at 10:40 AM
destevez.net
There is now Sentinel-2 imagery for today that has no clouds and has data products that can be downloaded for free. This shows a comparison between 2025-08-10 and 2025-08-13.
August 13, 2025 at 8:17 PM
destevez.net
The above image was supposed to be an animated GIF comparing before and after the fire, but it looks like that doesn't work on Bluesky. Here are the two images.
A satellite image on a cloudless day a few days before the fire. It is shown for comparison. A satellite image of the day after the fire. There are some clouds, but most of the burnt area can be seen through a break in the clouds.
August 13, 2025 at 10:02 AM
destevez.net
On Monday there was a large wildfire in Tres Cantos, Madrid, Spain, which is where I live (my family, friends and I are all okay). Today I have found some Airbus Space Pléiades Neo imagery that shows the extent of the fire more accurately than all other source I've seen.
An animated GIF that toggles between two satellite images of the same area. One was taken on a cloudless day some days before the fire. The other one is from the day after the fire. There are some clouds, but most of the burnt area can be seen through a break in the clouds.
August 13, 2025 at 10:00 AM
destevez.net
I've made new releases of gr-satellites: 5.8.0 (GNU Radio 3.10), 4.15.0 (GNU Radio 3.9), 3.22.0 (GNU Radio 3.8). This will be the last release supporting GNU Radio 3.9 and 3.8. Maintainance for these branches will stop now.
The changelog of the release: Added Added callsign to deframer options Basic support for SigMF input CSP Address Filter block CSP ZMQ blocks Frame counter block Mobitex-NX deframer Support for JINJUSAT-1B Support for Hex20 Nila Support for NUSHSAT-1 Support for TEVEL2 satellites Telemetry decoder for TUBiX20 (TUBIN) Time-dependent delay block Time synchronization tag to Doppler correction block U482C encoder block Uplink argument to tle_to_doppler_file.py example Fixed Callsign '-' handling in Check AX.25 Address Packet size bug in AX100 deframer pybind11 bug in swap_header Timezone handling in PDU to KISS Changed Added callsign in SatYAML to BEESAT-1,-9, TUBIN, NanoFF-A,-B, and TechnoSat Improved version printout in gr_satellites CLI app Updated AX.25 packet radio examples Use GNU Radio crc_check in GNU Radio >= 3.10
August 1, 2025 at 12:08 AM
destevez.net
New blog post: n78 band 5G NR recordings. I comment on some recordings I have made that are representative of the current state of 5G deployment on this band in Spain (100 MHz cells for Movistar and Orange, 90 MHz cell for Vodafone).
An inspectrum waterfall that shows a 100 MHz 5G cell that is rather busy. We can see 7 SSB repetitions, and mostly empty uplink slots, since this is a TDD band.
July 17, 2025 at 10:32 AM
destevez.net
Some months ago I shared my design of a PCB with an ADF4158 PLL/waveform synth and HMC319 VCO intended as an LO source for an FMCW radar in the 10 GHz band. I have finished soldering all the components, and the HMC319 appears to be working. The next step is the ADF4158 config.
A photo of a small PCB with an SMA connector connected to an attenuator, two leads to supply power, a green LED, and a bunch of surface mount components, including three QFN chips. A screenshot of GQRX showing a signal at 3.786 GHz with significant phase noise. This is the output of the VCO without the PLL being configured.
July 13, 2025 at 11:33 AM
destevez.net
New blog post: Galileo OSNMA chain renewal. OSNMA is the Galileo open service navigation message authentication system, which authenticates cryptographically the navigation data broadcast by the satellites. Here I comment on a TESLA chain renewal test done on Tuesday.
A diagram taken from the ICD that shows how the TESLA chain renewal happens in two steps. In the first step, the KROOTs for both chains are transmitted alternatively and the chain and pubkey status is end-of-chain. In the second step, the system transitions to using the new chain.
July 12, 2025 at 3:49 PM
destevez.net
I compare the legacy loop filter in GNU Radio with a correct implementation of the loop filter, showing that when the bandwidths are adjusted for similar convergence time, the correct loop filter has less noise.

Read more: destevez.net/2025/07/abou...
A plot of the value of the FLL frequency accumulator during lock-in. The legacy GNU Radio loop filter and the new fixed loop filter are compared. Both converge almost at the same time, but the legacy filter has more noise after convergence.
July 2, 2025 at 10:58 AM
destevez.net
I calculate the discriminant gain and show how to design loop filter coefficients properly, pointing out what is wrong with the current GNU Radio implementation of the loop filter.
A plot of the discriminant curves of the GNU Radio and Remez band-edge filters obtained via simulation. They are compared with the slope given by the theoretical gain at zero error.
July 2, 2025 at 10:58 AM
destevez.net
Then I compare ways of realizing band-edge filters as an FIR filter, including an idea by fred harris, which is what GNU Radio uses, and a Remez design.
A plot that compares the GNU Radio FIR filter, which extends the ideal quarter-sine band-edge to a half-sine, and a Remez design, which has more ripple but a narrower transition down to zero.
July 2, 2025 at 10:58 AM
destevez.net
New blog post: About FLLs with band-edge filters. This post is motivated by some bugs in the @gnuradio.org FLL that I've been working on fixing. I explain what band-edge filters are, and how they can be used in an FLL.
A plot of the spectrum of an RRC waveform and ideal band-edge filters. The band-edge filters are equal to the derivative of the RRC, so they are non-zero only in the roll-off region.
July 2, 2025 at 10:58 AM
destevez.net
New blog post: Z-Sat VHF transmissions. Recently Daniel Ekman has noticed that Z-Sat, a Japanese satellite launched in 2021 that has an amateur radio payload, is transmitting a digital signal continuously. Here I analyze a recording of this signal.
A screenshot of a GNU Radio GUI where we can see the soft symbols in the time domain and the output of a G3RUH descrambler. The descrambler output is all zeros.
June 30, 2025 at 7:12 PM
destevez.net
Finally I spend some time discussing some of the parameters that appear in the SIB1, which can be parsed with Python or in Wireshark.

Read more: destevez.net/2025/06/5g-n...
A screenshot in Wireshark that shows part of the SIB1 message parsed out. The PCAP file also contains the MIB transmission from the PBCH, which happen every 10 ms.
June 13, 2025 at 7:59 AM
destevez.net
I have added support to generate alists for the 5G LDPC codes to my ldpc-toolbox. I use this and a simple sum-product algorithm implementation to decode the PDSCH. I show how the LLRs evolve during the decoding iterations and how all the punctured bits get recovered.
A plot of the LLRs after the first LDPC decoder iteration. Many punctured bits are still not solved and have an LLR of zero. A plot of the LLRs after the second LDPC decoder iteration. Now all the punctured bits are solved and the codeword is correct.
June 13, 2025 at 7:59 AM
destevez.net
New blog post: 5G NR PDSCH. In this post I show how to decode the 5G PDSCH (downlink data) by using the SIB1 from an srsRAN gNB recording as an example. I go into the details of how LDPC and the rest of rate matching and other encoding work.
A waterfall of the IQ recording, which shows annotations marking the PDSCH, and other transmissions such as the PDCCH, the PSS, SSS and PBCH, and the CSI-RS.
June 13, 2025 at 7:59 AM
destevez.net
I explain some heuristics to find the appropriate DCI size. In particular, Polar decoding will likely fail if we assume a too small DCI size. By using this property and checking the CRC for a few DCI candidate sizes, we can find the correct size.
A plot of the output LLRs of Polar decoding for a DCI candidate size of 34. Some LLRs are very close to zero, which means that decoding has failed. A plot of the output LLRs of Polar decoding for a DCI candidate size of 35. We can see that all the LLRs are well away from zero, meaning that decoding has succeeded, and the correct DCI size is either this one, or slightly larger.
May 10, 2025 at 7:23 PM