LilyGO T-Deck and T-LoRa Pager: Drowning in Their Own Noise?
LilyGO T-Deck and T-LoRa Pager: Drowning in Their Own Noise?
A handheld can transmit a perfectly usable signal and still struggle to receive the reply. Our tests of a LilyGO T-Deck and T-LoRa Pager showed why receiver noise deserves as much attention as transmit power—and why the complete handheld matters more than the radio chip on its specification sheet.
The repeater hears you. Why can’t you hear the repeater?
That asymmetry is a useful reason to investigate the receive path. In these tests, our ON0RFG MeshCore repeater sent numbered messages, but reception on the handhelds was unreliable in the configurations we examined. Increasing handheld transmit power would not repair its reception of those messages.
A receiver has to extract a wanted signal from everything else at its input. Some of that unwanted energy comes from the surroundings. Some can come from the receiver’s own processor, regulators, display circuitry or attached cables. Put a sensitive radio and a busy little computer in the same enclosure, and keeping them from interfering with each other becomes part of the antenna-system design.
Self-interference means unwanted energy generated by the device reaches its own receiver. The resulting loss of weak-signal reception is often called desensitization. A good radio IC cannot make noise disappear after it has arrived alongside the wanted signal.
These are the results from our October 15, 2026 field report, using one 433 MHz T-Deck, one 433 MHz T-LoRa Pager and our RF.Guru pager. They describe those units and those test configurations. The useful finding is real: reception and reported noise changed substantially with the handheld and its setup. The measurements also tell us where to be careful about attributing the cause.
The important specification is whether the complete device receives reliably. Chip sensitivity, output power and a large display do not establish that. A short numbered-packet test, repeated with controlled changes, can reveal a problem that the feature list misses.
What we put on the bench
The repeater sent numbered messages on the #mesh channel every three seconds. The handhelds were tested at the same location, and we counted the messages that each decoded.
Scroll sideways to see every column.
| Station | Radio hardware | Antenna | Reported transmit power |
|---|---|---|---|
| ON0RFG repeater | NiceRF LoRa1268F30-T; SX1268 with temperature-compensated reference | Procom 0 dBd / 2.15 dBi vertical, 6 m above the roof, through a cavity filter | About 1 W |
| LilyGO T-LoRa Pager | SX1262, 433 MHz version | Built-in antenna | 22 dBm |
| LilyGO T-Deck | SX1262, 433 MHz version | LilyGO whip on SMA; RF.Guru stub in a separate run | 22 dBm |
| RF.Guru pager | RP2040 and RFM96W / SX1276 | 2 cm stub | About 17 dBm |
The channel settings were 434.890 MHz, 62.5 kHz bandwidth, spreading factor 8, coding rate 4/8 and a 16-symbol preamble. These are the settings of this amateur-radio experiment, not a general license-free operating recommendation.
Bandwidth is the width of spectrum the receiver processes. The spreading factor, abbreviated SF, sets part of the LoRa waveform’s timing and processing gain. Raising SF can make weaker signals decodable, at the cost of longer messages on air. It does not remove the interference.
The powers in the table describe the complete stations, not a controlled transmit-power comparison. For scale, 22 dBm is approximately 158 mW and 17 dBm is approximately 50 mW: a 5 dB difference, or roughly one-third the power. Handheld transmit power does not determine its reception of the repeater.
Take the messaging application out of the first comparison
We flashed the LilyGO units with a receive-only sniffer built around RadioLib. It did not transmit or run the normal display and audio application. It logged decoded packets, reported signal strength, signal-to-noise ratio, frequency error, packet-error events and periodic idle RSSI over USB.
This helps separate radio reception from the behavior of a full messaging application. It does not make the remaining processor, regulators, USB connection or peripheral supply rails disappear. No display activity is not the same as proving every display-related circuit has been electrically disconnected.
RSSI is the radio’s received-signal-strength indication. SNR is its estimate of signal-to-noise ratio. In this article, “noise floor” means the reported idle RSSI in that receiver configuration. These are useful diagnostic readings, not calibrated spectrum-analyzer measurements or laboratory noise-figure results.
The packet counts and reported levels
Scroll sideways to see the signal, SNR and packet-count columns.
| Device and antenna | Idle RSSI | Packet signal | Reported SNR | Decoded |
|---|---|---|---|---|
| T-LoRa Pager, built-in antenna | −84 dBm | −91 to −94 dBm | −8 to −11 dB | 7 / 20 |
| T-Deck, LilyGO whip | −98 dBm | −107 to −110 dBm | −10 to −13 dB | 14 / 20 |
| T-Deck, RF.Guru stub | −99 dBm | −111 dBm | −12.8 dB | 1 / 10 |
The Pager missed most messages in its run. The T-Deck with its whip decoded a majority, but still missed six out of twenty. With the stub it decoded only one of ten. Our RF.Guru pager decoded 30 of 30 at the test location in the reported comparison.
The idle and packet readings were not simultaneous calibrated measurements. Do not subtract the displayed columns and expect an exact independent reconstruction of the chipset’s SNR estimate. LoRa can decode at negative SNR, and its packet estimator is not simply the difference between two displayed RSSI values.
Missing messages produced no reported CRC or header-error events in these runs. That is consistent with packets failing before a recognizable receive event. It does not, by itself, prove a particular reason for every missed packet.
One result must stay beside the bad results: earlier the same day, the same T-Deck running MeshCore companion firmware decoded 30 of 30. The later sniffer runs used a USB-connected laptop. Position, orientation, cabling, firmware configuration and power arrangement were not all isolated variables. The T-Deck was not permanently incapable of receiving that path.
Changing the repeater and T-Deck to SF10 improved the stub configuration from 1/10 in the SF8 run to 14/20 in the SF10 run, with packets reported down to −17.8 dB SNR. That is a useful operating observation, not a measured universal sensitivity threshold. The price is more airtime, and the elevated noise still deserves investigation.
The 50-ohm load is the revealing experiment
We replaced the T-Deck antenna with a 50-ohm termination: a suitable load on the antenna socket that largely removes normal antenna pickup while keeping a defined input connection.
| Input connection | Idle RSSI |
|---|---|
| LilyGO whip | −98 dBm |
| RF.Guru stub | −99 dBm |
| 50-ohm load | −113 dBm |
That is a 15 dB reduction from the whip reading, or 14 dB from the stub reading. The receiver was seeing substantially more input-associated energy with an antenna connected. This is a much more useful clue than arguing over the sensitivity number in a chip brochure.
A plausible mechanism is the handheld radiating into its own nearby antenna. Digital edges, clock activity and switching supplies can couple into that path. However, the load also removes pickup of room noise and radiation from the laptop or USB cable. Changing the antenna changes its interaction with the enclosure and cable too. This test identifies an antenna-dependent contribution; it does not uniquely identify the source.
The fact that another pager worked at the same spot is valuable comparison evidence, but it does not eliminate ambient interference. Different antennas, patterns, receivers and attached cables can respond differently to the same field. A quiet reference receiver placed nearby is helpful; a controlled substitution test is stronger.
A T-Deck sweep between 433 and 436 MHz gave idle readings around −97 to −102 dBm. That argues against one isolated narrow spur being the whole problem in that setup. It does not guarantee that every channel, bandwidth or physical arrangement behaves identically.
What the RF.Guru pager comparison establishes
Our pager gave the better practical reception result in these field notes. It decoded 30/30, and the original report records continued two-way operation with a 20 dB attenuator in its antenna line, while using its short stub and lower transmit power.
That is a useful demonstration of operating margin in that particular configuration. It is not a calibrated “20 dB better receiver sensitivity” specification for the RF.Guru design, nor a 20 dB penalty that can be assigned to every LilyGO handheld. The antennas and hardware were different; the source does not record an identical swept-attenuation packet-error test for all three units. Adding attenuation can also reduce interference picked up by an antenna.
Our design aim is straightforward: keep the radio input and antenna environment quiet, including the processor, display, power system and connections. The results support paying attention to that complete implementation. They do not isolate the RP2040 versus ESP32-S3, the display technology or the radio IC as the sole cause of the difference.
How to investigate your own handheld
Begin with reception you can count. Use a known transmitter sending numbered messages, and record sent and decoded totals. Keep frequency, bandwidth, SF, coding rate, preamble, transmit power, position and antenna orientation fixed. Save the exact firmware versions and settings. Repeat an A–B–A sequence so a change in propagation does not quietly become a claimed hardware improvement.
Start on battery with unnecessary cables removed. Compare USB disconnected and connected without moving the handheld. Then vary laptop or charger power, cable routing, display activity and other peripheral activity one at a time. If logging requires a cable, that cable is part of the test configuration; record it rather than treating it as invisible.
On a device with a suitable external antenna connector, compare the antenna with a properly connected 50-ohm termination. Then try the same antenna farther from the handheld using a known feedline, keeping orientation and surroundings controlled as far as possible. Record the cable’s loss: a lower noise reading after adding cable is not automatically evidence that the noise source has vanished.
The T-LoRa Pager’s built-in antenna did not permit the same simple substitution in our test. Do not pretend that its −84 dBm reading came from an identical antenna-port experiment. For any hardware modification, preserve a reproducible baseline and avoid drawing conclusions from a changed antenna match alone.
For a genuine receiver-sensitivity comparison, use a known LoRa signal, a shielded conducted setup where practical, calibrated attenuation and a declared packet-error criterion. Sweep the signal level and use longer repeated packet runs. Compare all units at the same reference plane and settings, separating antenna performance from receiver performance. Idle RSSI alone is not that measurement.
What we would change before adding transmit power
First establish a quiet receive configuration: battery operation, sensible cable placement, controlled antenna position and verified firmware behavior. If a particular display or power mode consistently raises idle RSSI and reduces packet reception, fix that path before paying for more output power.
The original notes also report a separate pause in reception after messages with Ripple firmware. That deserves its own timing investigation. A sniffer removes that application behavior from the test, but changing firmware can also change hardware activity. Separate packet-processing gaps from RF desensitization instead of explaining both with one symptom.
The field result is useful precisely because it is concrete. These two LilyGO units showed poor or inconsistent reception in the documented runs; the RF.Guru pager worked well at the same location. The T-Deck’s 14–15 dB antenna-to-load difference points directly toward an input-associated noise problem worth tracing. The next step is to isolate that path, not declare a whole brand deaf or assume that more transmit watts will make a receiver listen better.
Mini-FAQ
- Does this mean every LilyGO T-Deck or T-LoRa Pager has the same problem? No. The report covers one 433 MHz unit of each model, one location and short test runs. Earlier that day the same T-Deck decoded 30/30 with a different operating configuration.
- Why can the repeater hear me while I miss its replies? The two receive paths have different antennas, noise environments and receiver implementations. More handheld transmit power does not lower noise at the handheld’s input.
- Does the 50-ohm load prove the ESP32 causes the noise? No. The 14–15 dB drop demonstrates an antenna-dependent contribution. It does not separate the handheld’s emissions from ambient or attached-cable interference, or identify one component.
- Does the RF.Guru pager have a measured 20 dB sensitivity advantage? This field comparison does not establish that specification. It records successful operation with 20 dB added attenuation in that pager’s setup, without an identical conducted threshold test across all devices.
- Can a higher spreading factor help? It helped the reported T-Deck stub configuration: 1/10 at SF8 became 14/20 in the SF10 run. Higher SF costs airtime and does not remove the underlying noise.
- Where are the original notes and test firmware? The RF.Guru MeshCore repeater repository contains the October 15 field report and the receive-only rx-sniffer source. Both are linked below.
Field notes and reproducible tools
Original RF.Guru field report · Receive-only sniffer source · LilyGO hardware library and board definitions · MeshCore T-Deck variant
The original runs were short and used radio-reported levels, not calibrated laboratory measurements. No raw packet-log archive is supplied with this article. The hardware, antennas, firmware, cabling and sample counts above define the scope of the comparison.
Found this useful? Support my writing and experiments: Buy me a coffee.