A drone navigation board is a GNSS receiver bolted to a vibrating, sun-exposed airframe that shares its power rail with four motors. The reference clock on that board is not really chosen by frequency. It is chosen by four things that the datasheet headline never mentions.
Start with what the board will not give you. There is no spare area near the GNSS front end, because that patch of copper is the quietest real estate on the PCB and the reference has to sit inside it, between the receiver IC and the antenna feed. There is no spare current either, since the RF chain runs for the whole flight and the reference is the one rail you cannot put to sleep mid-air. And there is no thermal margin worth guessing at, because a sealed pod in direct sun, or a board sitting a centimetre from an ESC heatsink, runs well above the airframe's ambient temperature. Add supply lifetime on top, because a drone product ships for years.
The Epson X1G0054210405 is a 32 MHz TCXO from the TG2520SMN family, and it clears most of that list outright. It clears the size constraint at 2.5 × 2.0 × 0.8 mm. It clears the supply constraint with a 1.7 V to 3.63 V range, so it can hang straight off a 1.8 V or 3.3 V rail. It clears the drift constraint with ±0.5 × 10⁻⁶ maximum first-year aging at 32 MHz, and it clears the power-up constraint with a 1.0 ms maximum start-up time. Where it goes marginal is temperature and noise. The datasheet specifies only two temperature-stability grades, both pinned to -40 °C to +85 °C, and it publishes no phase-noise table at all. If your receiver vendor wants a dBc/Hz number, or your enclosure runs hotter than +85 °C, stop here.
The constraints that actually decide this
Drone navigation hardware is unforgiving for the reference oscillator, because the board is small, the airframe moves and the enclosure breathes hot air.
Footprint and height. A 2520 body occupies 2.5 mm by 2.0 mm and stands 0.8 mm tall. On a board that also carries the flight controller, the barometer, the magnetometer and increasingly the IMU, that is the difference between fitting the RF keep-out zone and crowding it. The height matters as much as the area: low-profile pods and stacked boards rarely have more than a millimetre of clearance.
Continuous current. A GNSS receiver cannot stop tracking during a flight without paying for it in re-acquisition. The oscillator that feeds it runs continuously, so its current is a fixed entry in the power budget rather than a duty-cycled one. At 32 MHz, the datasheet ceiling is 1.8 mA.
Board temperature, not air temperature. This is the constraint that fails quietly. The part's operating window is -40 °C to +85 °C, and its storage window is barely wider at -40 °C to +90 °C. A pod that sits in direct sunlight with no airflow, or a board mounted close to the ESC power stage, can push the part's own footprint past the ambient the airframe reports. The thermocouple goes on the oscillator footprint, not on the wing.
Stability and aging. A tight reference buys loop margin for carrier tracking, and aging sets the drift that no factory calibration removes.
Supply lifetime. Package, pinout and ordering code have to stay stable across a production run measured in years.
Those constraints do not leave much room. What survives is a small-body, wide-supply oscillator with a documented temperature characteristic and a documented aging figure.
Mapping each constraint to a datasheet number
Every number below comes from the TG2016SMN / TG2520SMN specification sheet. The left column names the constraint the number answers.
| Constraint | Item / symbol | Datasheet value | Conditions on the sheet |
|---|---|---|---|
| Footprint and height | External dimensions, TG2520SMN | 2.5 × 2.0 × 0.8 mm | Body size 2.5 × 2.0 mm |
| Frequency plan | Output frequency range, fo | 10 MHz to 55 MHz | 32 MHz is one of 17 listed standard frequencies, from 16 MHz to 52 MHz |
| Rail compatibility | Supply voltage, VCC | 1.7 V to 3.63 V range | Regulated options 1.8 V ±0.1 V, 2.8 V ±5 %, 3.0 V ±5 %, 3.3 V ±5 % |
| Board temperature | Operating temperature range, T_use | G, -40 °C to +85 °C | Only grade offered on this sheet |
| Board temperature | Storage temperature range, T_stg | -40 °C to +90 °C | Storage as single product |
| Initial accuracy | Frequency tolerance, f_tol | ±1.5 × 10⁻⁶ Max. | After reflow, +25 °C |
| Stability grade | Frequency/temperature characteristics, fo-TC | C, ±0.5 × 10⁻⁶ Max. / F, ±2.0 × 10⁻⁶ Max. | Both over -40 °C to +85 °C; F is the standard stability version |
| Load pulling | Frequency/load coefficient, fo-Load | ±0.1 × 10⁻⁶ Max. | 10 kΩ // 10 pF ±10 % |
| Rail noise | Frequency/voltage coefficient, fo-VCC | ±0.1 × 10⁻⁶ Max. | VCC ±5 % |
| Lifetime drift | Frequency aging, f_age | ±0.5 × 10⁻⁶ Max. | +25 °C, first year, applies for 24 MHz ≤ fo ≤ 40 MHz |
| Power budget | Current consumption, ICC | 1.8 mA Max. | 26 MHz < fo ≤ 40 MHz |
| Receiver input | Output voltage, Vpp | 0.8 V Min. | Peak to peak |
| External parts | Output load, Load_R / Load_C | 10 kΩ // 10 pF | DC cut capacitor = 0.01 µF |
| Power-up | Start-up time, t_str | 1.0 ms Max. | t = 0 at 90 % VCC |
Read the table as separate arguments rather than one feature list.
The TG2520SMN body is 2.5 × 2.0 × 0.8 mm, which is the larger of the two bodies this sheet covers and the one that matches the 2520 designation. The same sheet also describes the smaller TG2016SMN, and the pin map and electrical table are shared.
The rail headline is the range itself: 1.7 V to 3.63 V is wide enough that the part needs no regulator of its own, which saves space on a board where every square millimetre near the receiver is contested. The 1.8 V, 2.8 V, 3.0 V and 3.3 V options are set by the ordering symbol, not by an external divider.
On drift, 32 MHz lands in the 24 MHz to 40 MHz band, so first-year aging is ±0.5 × 10⁻⁶ maximum. The sheet allows a looser ±1.5 × 10⁻⁶ above 40 MHz and from 20 MHz to 24 MHz, so check where your frequency sits before copying the number across.
Load and rail sensitivity are both ±0.1 × 10⁻⁶ maximum, so the reference will not punish a rail that moves within ±5 %, and it will not pull when the receiver's input impedance shifts.
Power-up is 1.0 ms maximum from 90 % VCC, fast enough that the reference is not the slowest element in a cold-start chain. Note the 90 % threshold when you check it on the bench.
Where the part stops being the right answer
Three places, and all three are readable straight off the sheet.
The first is noise. The datasheet's feature line says "low noise," and that is the end of it. No L(f) table, no dBc/Hz figure at any offset, no phase-noise plot. If your receiver vendor publishes a required phase-noise mask, this document will not let you close that check. The Epson TG-5006CJ family in the same catalogue does print one, running from -85 dBc/Hz at 10 Hz offset to -148 dBc/Hz at 10 kHz and beyond, so Epson specifies the parameter when it wants to. Here it is simply absent, which is not the same as bad, but is the same as unverifiable.
The second is the temperature ceiling. -40 °C to +85 °C is the only operating range on the sheet, and storage is -40 °C to +90 °C. There is no +105 °C or +125 °C option to step up to. For a pod-mounted GNSS board with airflow this is comfortable. For a board sitting next to the power stage of a high-current airframe, it is worth a thermal measurement before you commit, because the failure mode here is not a dead part, it is a part that drifts further than your error budget assumed.
The third is the stability grade. The datasheet defines both grades, C at ±0.5 × 10⁻⁶ and F at ±2.0 × 10⁻⁶ over -40 °C to +85 °C, and F is called the standard stability version. The grade is carried in the ordering code, so the exact part number you buy decides which one you get. If the code resolves to F, the oscillator becomes the largest single contributor in your frequency error budget, and everything else in this table is a rounding error next to it.
There is a fourth boundary that is not a number. The output is a clipped sine wave, not a CMOS square. It needs a 0.01 µF DC-cut capacitor at the output pin and a 10 kΩ // 10 pF load, and the receiver's reference input has to contain a squaring stage. Plenty of GNSS front ends do. If yours expects a square-wave clock, you are adding a gate, and that gate has its own jitter.
Bring-up checklist
- Decode the ordering code before layout and confirm which stability grade the exact part number carries. Both grades are on the same sheet and they are a factor of four apart.
- Confirm the ④ supply symbol matches the rail you actually have. Grade E targets 1.8 V, and the 2.8 V, 3.0 V and 3.3 V options are separate symbols.
- Place the bypass capacitor between VCC and GND, 0.01 µF to 0.1 µF, as close to the part as the layout allows. The datasheet asks for this explicitly for stable operation.
- Add the 0.01 µF DC-cut capacitor at the OUT pin. It is not inside the package, and the output load is specified as 10 kΩ // 10 pF.
- Keep pin 1 open or tied to GND on the TCXO build. It is an N.C. pin and it does not work as a ground pin.
- Check peak-to-peak at the receiver input, not at the oscillator pin. The minimum is 0.8 V, and the DC-cut capacitor sits between the two measurement points.
- Verify the supply stays inside 1.7 V to 3.63 V across the envelope, including motor transients and battery sag at the end of a flight.
- Measure temperature at the oscillator footprint with the airframe assembled and running, and time the cold start against the 1.0 ms maximum.
Frequently asked questions
Does a 2.5 × 2.0 mm TCXO really save board area?
Only relative to what it replaces, and this part is the larger of the two bodies on the same specification sheet. The TG2016SMN is 2.0 × 1.6 × 0.73 mm, and the electrical table, pin map and output waveform are shared with the TG2520SMN. The 2520 body buys the same behaviour with more room between the pads, which matters where DFM rules near the RF section are already tight.
Is -40 to +85 °C enough for a drone?
For a pod with airflow, almost always. For a board mounted near the ESC power stage, or a sealed enclosure sitting in direct sun before take-off, it depends on your measured board temperature rather than the ambient. Note that storage is only -40 °C to +90 °C. If the oscillator footprint goes past +85 °C, this family has no wider grade to step up to.
Do I need external components around it?
Two, and both are cheap. A bypass capacitor of 0.01 µF to 0.1 µF between VCC and GND, as close to the part as the layout allows, is called out on the datasheet. A 0.01 µF DC-cut capacitor on the output completes the specified 10 kΩ // 10 pF load. There are no external load capacitors to tune, because the compensation is inside the package.
How does the supply side look over a multi-year build?
The series spans 10 MHz to 55 MHz, and 32 MHz sits in the middle of that range rather than at an edge, which usually helps long-term availability. X1G0041310006, X1G0041310009, X1G0041310010, X1G0041310011 and X1G0041310014 show how Epson splits the same space: those TG-5006CJ parts are 16.367667 MHz, 19.2 MHz, 26 MHz, 38.4 MHz and 16.368 MHz in the smaller 2016 body, from a fixed 2.85 V to 3.15 V rail over -30 °C to +85 °C, with ±0.5 × 10⁻⁶ or ±2.0 × 10⁻⁶ temperature grades and a published phase-noise table. The trade is consistent: the siblings give up the wide supply range and 10 °C at the cold end for a smaller body and a specified noise floor. Confirm the ordering code before freezing the BOM.
Sourcing note
The X1G0054210405 ships through Mandu with the full Epson specification sheet. Send us your rail voltage, your measured board temperature and your annual volume, and we will confirm the stability grade carried by the exact part number before you release the layout.