WSPR PROTOCOL

Configuration, Operation & Message Format

A Technical Reference for the GCARC WSPR Network · W2MMD · Spring 2026

1. Overview

WSPR (Weak Signal Propagation Reporter, pronounced “whisper”) is a narrowband digital radio protocol designed by Nobel Prize laureate Joe Taylor (K1JT) specifically for automated, unattended propagation beacon operation at the lowest possible power levels. The protocol was first released in 2008 and has grown into a worldwide network of thousands of continuously operating beacons and receive stations.

The core objective of WSPR is simple: encode the minimum information needed to identify a transmitting station and its location, then transmit that information in a way that can be decoded reliably even when the received signal is far below the audible noise floor. This is achieved through a combination of very slow transmission speed, continuous-phase 4-tone frequency-shift keying, and a powerful forward error-correcting code.

A WSPR beacon transmits 200 milliwatts every ten minutes, around the clock, with no operator intervention. Receiving stations around the world decode the signal, log the spot, and upload it automatically to public databases. The result is a real-time, global map of which propagation paths are open — updated every two minutes, 24 hours a day.

2. Operating Frequencies

2.1  Dial Frequency vs. Transmit Frequency

WSPR uses USB (Upper Sideband) mode on a conventional SSB transceiver or, in the case of a Pi Zero / TAPR HAT system, generates the RF signal directly. In the SSB transceiver case, the radio is set to a specific “dial frequency” in USB mode, and the WSPR audio tones are injected at approximately 1,400–1,600 Hz audio offset, placing the actual RF transmission 1,500 Hz above the dial frequency.

The WSPR sub-band on each HF frequency is only 200 Hz wide, centered 1,500 Hz above the dial frequency. Multiple stations can coexist within this 200 Hz window because each beacon picks a random transmit frequency within the window on each transmission cycle.

Band	USB Dial Freq (MHz)	TX Center (MHz)	TX Range	Notes
2190m	0.136000	0.137500	±100 Hz	LF — special license required
630m	0.474200	0.475700	±100 Hz	MF — Part 97.303 in USA
160m	1.836600	1.838100	±100 Hz	NVIS, night only
80m	3.568600	3.570100	±100 Hz	Regional NVIS, night DX
60m	5.287200	5.288700	±100 Hz	Check local allocation
40m	7.038600	7.040100	±100 Hz	Day + night, popular band
30m ★	10.138700	10.140200	±100 Hz	RECOMMENDED — most active WSPR band
20m	14.095600	14.097100	±100 Hz	Best global DX range
17m	18.104600	18.106100	±100 Hz	WARC — no contests
15m	21.094600	21.096100	±100 Hz	Solar-dependent DX
12m	24.924600	24.926100	±100 Hz	WARC — no contests
10m	28.124600	28.126100	±100 Hz	Technician OK — outstanding DX when open
6m	50.293000	50.294500	±100 Hz	Sporadic-E / Technician OK
2m	144.489000	144.490500	±100 Hz	Mostly local/aircraft scatter
70cm	432.300000	432.301500	±100 Hz	Local only
23cm	1296.500	1296.501	±100 Hz	Microwave experimental

★ 30m note: The 30m band (10.140 MHz TX center) is the most active WSPR band worldwide. It is a WARC band, meaning no contest operation, no phone — ideal for a 24/7 unattended beacon. For GCARC members, the TAPR HAT dial frequency is 10.1387 MHz.

2.2  The WSPR Sub-Band

On any given band, all WSPR activity is confined to a 200 Hz window centered 1,500 Hz above the dial frequency. Because each WSPR signal occupies only about 6 Hz of bandwidth, roughly 30 stations can simultaneously occupy the same 200 Hz window without mutual interference. The transmitter software picks a random audio offset within 1,400–1,600 Hz on each transmission, spreading stations evenly across the window.

On the TAPR HAT / Pi Zero 2W system, the Pi generates the RF signal directly (no SSB transceiver required). The WsprryPi software generates the four FSK tones at the correct RF frequency, eliminating the audio chain entirely. The configured frequency is the actual transmitted RF frequency, not a dial offset.

2.3  Frequency Accuracy Requirements

WSPR decoding requires the transmitter frequency to be accurate to within a few Hz. The protocol can tolerate some frequency error because the decoder searches a window, but excessive drift causes missed decodes. The Pi Zero 2W’s onboard crystal oscillator is adequate for most WSPR operation; the WsprryPi software includes a PPM (parts per million) correction parameter to compensate for individual crystal error.

A 1 PPM error at 10 MHz = 10 Hz of frequency offset. Typical Pi Zero crystals are within ±1–2 PPM, so frequency error is generally under 20 Hz — well within the decoder’s tolerance.

3. Timing and Transmission Schedule

3.1  The Two-Minute Timeslot

WSPR divides time into two-minute slots synchronized to UTC. All WSPR activity worldwide operates on this same grid: slots begin at 00:00, 00:02, 00:04 UTC, and so on throughout the day. Every WSPR operator, worldwide, on any band, is synchronized to the same timeslot boundaries.

Within each two-minute slot, the sequence is:

• T = 0:00 — Start of even UTC minute. All stations begin listening.
• T = 0:01 — Transmissions begin exactly 1 second after the even minute.
• T = 0:01 to 1:51 — Transmission in progress (110.6 seconds of continuous 4-FSK tones).
• T = 1:52 to 2:00 — Transmission ends. Decoder processes the received audio.

The 1-second offset at the start of the slot prevents transmitters from stepping on the very start of the receive window, giving the decoder a clean edge to lock onto.

3.2  Duty Cycle — How Often a Station Transmits

A WSPR station does not transmit every two-minute slot. The standard duty cycle is 20% — meaning one transmission in every five slots, on average. Most operators also configure their station to listen (receive and decode) in the remaining slots, contributing spot data back to the network.

The transmit/receive decision is made pseudo-randomly each slot. The transmit percentage is configurable — common settings are 20%, 25%, or 33%. For a pure beacon (transmit only, no receive), 100% is permitted but discouraged as it contributes nothing to the network’s receive data.

Example for a 20% duty cycle station on 30m:
00:00 UTC — Listen
00:02 UTC — Listen
00:04 UTC — TRANSMIT (110.6 sec starting at 00:04:01)
00:06 UTC — Listen
00:08 UTC — Listen
00:10 UTC — TRANSMIT
  … and so on, randomly distributed at 20% rate

3.3  Clock Synchronization Requirement

Clock accuracy is critical. The receiver decoder uses the timing of the incoming signal to decode it, and the decode algorithm will fail if the transmitter’s clock is off by more than about ±1 second from UTC. The Pi Zero 2W synchronizes its clock via NTP (Network Time Protocol) over Wi-Fi. Internet connectivity must be maintained at all times during WSPR operation.

NTP on a well-connected Pi is typically accurate to within 10–50 milliseconds — far better than the ±1 second tolerance required. In practice, clock accuracy is rarely a problem with a Pi that has reliable Wi-Fi access.

4. Modulation — Continuous Phase 4-FSK

4.1  Four-Tone FSK

WSPR uses 4-tone frequency-shift keying (4-FSK), meaning the transmitter shifts between exactly four discrete frequencies to encode data. The four tones are spaced 1.4648 Hz apart (12000/8192 Hz, derived from the audio sample rate). The total occupied bandwidth of the four tones is about 6 Hz — extraordinarily narrow.

Each tone represents a 2-bit symbol (values 0, 1, 2, or 3). The modulation is “continuous phase,” meaning the transmitter never abruptly switches frequency mid-cycle; instead it smoothly transitions from one tone to the next. This suppresses splatter and keeps the signal within its 6 Hz footprint even at symbol transitions.

Parameter	Value	Derivation / Notes
Tone spacing	1.4648 Hz	12000 ÷ 8192 samples per symbol
Symbol duration	0.6827 seconds	Reciprocal of tone spacing (1 ÷ 1.4648)
Total symbols	162	(50 + 32 − 1) × 2 = 162 channel symbols
Total TX duration	110.6 seconds	162 symbols × 0.6827 s/symbol
Total bandwidth	~6 Hz	4 tones × 1.4648 Hz spacing ≈ 4.4 Hz + guard
Baud rate	1.46 baud	Symbols per second
Bits per symbol	2 bits	log₂(4) = 2
Gross bit rate	2.92 bits/sec	1.46 baud × 2 bits/symbol
Effective info rate	~0.45 bits/sec	50 info bits ÷ 110.6 sec

4.2  Why 4-FSK Instead of Binary?

Using 4-FSK instead of 2-FSK (binary) doubles the number of bits per symbol without increasing the bandwidth or symbol rate. This improves spectral efficiency. The choice of 4-FSK also simplifies the synchronization task for the decoder — the sync vector (see Section 6) is interleaved into the symbol stream as the most significant bit of each 2-bit symbol, with the data occupying the least significant bit.

5. Message Content — What Is Transmitted

5.1  The Standard WSPR Message

Every standard WSPR transmission carries exactly three pieces of information:

• Callsign — the transmitting station’s amateur radio callsign (up to 6 characters)
• Grid locator — a 4-character Maidenhead grid square identifying the transmitter’s location to within about 100 km × 110 km
• Power — the transmitter’s output power in dBm (0 to 60 dBm, in 3 dB steps)

Example message:

WB2MNF FN20 23

WB2MNF is the callsign. FN20 is the Maidenhead grid square for Mullica Hill, NJ. 23 is 23 dBm = 200 mW (the TAPR HAT’s typical output).

5.2  The Maidenhead Grid Locator

The Maidenhead Locator System divides the world into a hierarchy of grid squares. The 4-character locator (e.g., FN20) identifies a rectangle approximately 2° longitude × 1° latitude (about 111 km × 111 km in the mid-latitudes). The first two characters are letters (field), the second two are digits (square).

WSPR encodes only the 4-character locator. A 6-character locator (e.g., FN20qe) provides ~5 km precision but requires the extended 2-transmission sequence (see Section 5.4).

Locator Chars	Grid Size	Precision	Example
2 (field)	20° × 10°	~1,000 km	FN
4 (square)	2° × 1°	~111 km	FN20
6 (subsquare)	5′ × 2.5′	~5 km	FN20qe
8 (extended)	30″ × 15″	~500 m	FN20qe22

5.3  Power Encoding

Transmit power is reported in dBm. The WSPR protocol only accepts power values that end in 0, 3, or 7 (corresponding to the 3 dB steps of the standard power scale). Valid values are 0, 3, 7, 10, 13, 17, 20, 23, 27, 30, 33, 37, 40, 43, 47, 50, 53, 57, 60 dBm. Other values are rounded to the nearest valid step.

dBm	Milliwatts	Typical Use
0	1 mW	Minimum — ultra-QRP
3	2 mW	Ultra-QRP
7	5 mW	QRP
10	10 mW	QRP
13	20 mW	QRP
17	50 mW	QRP
20	100 mW	TAPR HAT minimum
23	200 mW	TAPR HAT typical output ★
27	500 mW	QRP transceiver
30	1 W	QRP transceiver
37	5 W	Standard QRP limit
43	20 W	Low power
50	100 W	Typical HF station
60	1,000 W	Maximum

5.4  Extended Messages — Two-Transmission Sequence

Callsigns that don’t conform to the standard format (e.g., compound callsigns like W1AW/4, or callsigns with prefixes like OE/WB2MNF) and stations that want to report a 6-character grid square use a two-transmission sequence. The first transmission uses the standard format; the second carries the extended information. Decoders that receive both transmissions in sequence can reconstruct the full data.

For the GCARC WSPR Network, all members will use standard-format callsigns and 4-character grid squares, so the single-transmission format applies to all beacons in this project.

6. Message Encoding — From Text to Symbols

The journey from the text string “WB2MNF FN20 23” to the 162 RF symbols transmitted over the air involves five distinct steps. This section describes each step.

6.1  Step 1: Source Compression — 50 Bits

The three fields (callsign, grid, power) are compressed into exactly 50 bits using a scheme that exploits the restricted character sets in each field:

Callsign Encoding (28 bits)

The callsign is normalized to exactly 6 characters:

• If the callsign is shorter than 6 characters and the second character is a digit (standard format), pad with spaces at the end.
• If the first character is a letter but the second is not a digit (e.g., “G4JNT”), prepend a space to make the third character a digit.

The 37 valid characters are assigned numeric values: digits 0–9 → 0–9, letters A–Z → 10–35, space → 36.

The 6-character callsign encodes as:

N = (c1 × 36 × 10 × 27 × 27 × 27) + (c2 × 10 × 27 × 27 × 27) + (c3 × 27³) + (c4 × 27²) + (c5 × 27) + c6

  where c1 must be 0–36 (any), c2 must be 0–36, c3 must be 0–9 (digit only),

  and c4–c6 must be 0–26 (letter or space only)

  Result: N fits in 28 bits (max value = 37 × 36 × 10 × 27³ – 1 ≈ 262,177,560)

Example: “WB2MNF” → W=32, B=11, 2=2, M=22, N=23, F=15

N = 32×36×10×27³ + 11×10×27³ + 2×27³ + 22×27² + 23×27 + 15

Grid Locator Encoding (15 bits)

The 4-character Maidenhead locator is encoded as:

M = (179 – 10 × (c1 – ‘A’) – (c3 – ‘0’)) × 180 + 10 × (c2 – ‘A’) + (c4 – ‘0’)

  Result: M fits in 15 bits (max = 32400)

Example: “FN20” → F=5, N=13, 2=2, 0=0

M = (179 – 10×5 – 2) × 180 + 10×13 + 0 = 127 × 180 + 130 = 22,990

Power Encoding (7 bits)

Power in dBm is encoded as:

P = power_dBm + 64

  Example: 23 dBm → P = 23 + 64 = 87  (fits in 7 bits, range 0–127)

Combined 50-bit Message

The three values are concatenated into a single 50-bit integer:

message_50bit = (N << 22) | (M << 7) | P

  N occupies bits 49–22 (28 bits)

  M occupies bits 21–7  (15 bits)

  P occupies bits 6–0   (7 bits)

6.2  Step 2: Convolutional Forward Error Correction (FEC)

The 50-bit compressed message is fed into a rate-1/2 convolutional encoder with constraint length K=32. “Rate 1/2” means for every 1 input bit, 2 output bits are produced. “Constraint length 32” means the encoder looks at the current input bit plus the previous 31 bits when computing each output pair.

Input: 50 information bits + (K-1=31) tail bits of zeros = 81 input bits.

Output: 81 × 2 = 162 coded bits.

The two generator polynomials that define the convolutional code are (in octal):

G1 = 0xF2D05351  (hex) = 11110010110100000101001101010001 (binary)

G2 = 0xE4613C47  (hex) = 11100100011000010011110001000111 (binary)

The long constraint length (K=32) provides very strong error correction. It makes undetected decoding errors extremely unlikely, at the cost that the efficient Viterbi algorithm can’t be used — the decoder must use a sequential (Fano) algorithm instead. In practice the WSPR decoder achieves reliable decoding at SNRs as low as −28 dB in a 2,500 Hz reference bandwidth.

6.3  Step 3: Interleaving (Bit Reversal)

The 162 coded bits are reordered (interleaved) to spread any burst of interference across non-contiguous bit positions, making the error correction more effective. WSPR uses “bit-reversed addressing”: the bit at index I is moved to position J, where J is the 8-bit reversal of I.

For example: bit at index 1 (binary 00000001) moves to index 128 (binary 10000000). Bit at index 13 (00001101) moves to index 176 (10110000).

This complete permutation of 162 bits is a fixed, deterministic mapping — the same lookup table is used by every WSPR encoder and decoder in the world.

6.4  Step 4: Synchronization Vector

WSPR uses a fixed 162-element synchronization (sync) vector — a pseudo-random binary sequence with good autocorrelation properties. The sync vector is interleaved with the data bits to form the final 162-symbol stream. Each symbol is 2 bits wide:

symbol[i] = 2 × sync[i] + data[i]

  sync[i]  → most significant bit (0 or 1)

  data[i]  → least significant bit (0 or 1)

  Result:  symbol values of 0, 1, 2, or 3

The sync vector is constant — it never changes. This allows the decoder to search for a WSPR signal by correlating the received symbol sequence against the known sync pattern, even before knowing the message content. The sync provides both signal detection and fine timing alignment.

The sync vector begins:

1,1,0,0,0,0,0,0,1,0,0,0,1,1,1,0,0,0,1,0,0,1,0,1,1,1,1,0,0,0,0,0,…

(162 values total — see WSPR source code for complete vector)

6.5  Step 5: Transmission

The 162 symbols (values 0–3) are transmitted as 4-FSK tones:

Symbol Value	Tone Offset from Carrier
0	0.0 Hz   (base tone)
1	+1.4648 Hz
2	+2.9297 Hz
3	+4.3945 Hz

Each tone lasts exactly 8192/12000 = 0.6827 seconds. Transitions between tones are continuous-phase (no abrupt jumps). The transmitter begins exactly 1 second after the even UTC minute and transmits all 162 symbols continuously, completing in 110.6 seconds.

7. The Receiver and Decoder

7.1  What the Receiver Hears

A WSPR signal sounds like a nearly continuous tone with a very slight, slow warble. The 4-FSK modulation shifts the frequency by at most 4.4 Hz over the entire 110-second transmission. At typical SNRs, the signal is completely inaudible — buried in band noise. Software decoders extract the signal by performing a detailed spectral analysis of the received audio.

7.2  Decoder Operation

At the end of each 2-minute slot, the WSPR decoder processes approximately 114 seconds of recorded audio (the 110.6-second transmission plus guard time). The steps are:

• FFT analysis — compute a high-resolution frequency spectrum with ~0.37 Hz resolution across the 200 Hz WSPR sub-band
• Sync correlation — slide the known sync vector across all frequency offsets and time offsets to find candidate signals
• Symbol extraction — for each candidate, extract the 162 symbol values from the 4 spectral bins at each symbol time
• FEC decoding — reverse the interleaving and run the sequential (Fano) convolutional decoder to recover the 50 information bits
• Message reconstruction — decode callsign, grid, and power from the 50-bit payload
• Database upload — the spot (callsign, grid, power, frequency, SNR, drift, receiving station) is uploaded to wspr.live / wsprnet.org

7.3  The SNR Figure

The decoder reports the received SNR in dB, referenced to a 2,500 Hz noise bandwidth. This is the “received SNR in 2.5 kHz” figure seen in all WSPR spot databases. Typical values range from about +10 dB (strong local signal) to −28 dB (the theoretical decoding limit). Values below −30 dB are occasionally decoded under exceptional conditions.

Since the WSPR signal occupies only ~6 Hz but the SNR is referenced to 2,500 Hz, there is an inherent “processing gain” of 10 × log₂(2500/6) ≈ +26 dB built into the system. This is why a 200 mW beacon can be decoded when it is far below audibility.

7.4  The Drift Parameter

The decoder also reports the frequency drift of the received signal in Hz/minute. A well-calibrated, thermally stable transmitter should show drift close to 0. Large drift values (>2 Hz/min) indicate oscillator instability — common with uncompensated crystal oscillators in very warm or cold environments. The Pi Zero’s crystal is generally stable enough for WSPR, but enclosures that trap heat can cause drift issues in summer months.

8. Spot Data — What Gets Logged

Every time a receiving station successfully decodes a WSPR signal, it logs a “spot” to the public databases. A spot record contains:

Field	Example	Description
timestamp	2026-03-21 14:04:00	UTC time of the 2-minute slot start
tx_sign	WB2MNF	Transmitting station callsign
tx_loc	FN20	Transmitting station grid square
tx_lat	39.75	Grid center latitude (derived)
tx_lon	-75.25	Grid center longitude (derived)
power	23	Reported TX power in dBm
rx_sign	K2ABC	Receiving station callsign
rx_loc	EL96	Receiving station grid square
rx_lat	26.5	Receiving station latitude (derived)
rx_lon	-80.5	Receiving station longitude (derived)
distance	1,842	Path distance in km (computed)
azimuth	197	Bearing TX→RX in degrees
snr	-18	Received SNR in dB re 2500 Hz BW
freq	10.140152	Actual received frequency in MHz
drift	0	Frequency drift in Hz/min
band	30	Nominal band in meters
version	2.6.1	Decoder software version

The wspr.live ClickHouse database stores all of these fields and is queryable via a free public SQL API at https://db1.wspr.live/ — no account required. The wspr.rocks website provides a browser-based SQL query interface on top of the same database.

9. WsprryPi Configuration Reference

For GCARC members using the TAPR HAT + Pi Zero 2W, configuration is done through the WsprryPi web interface at http://[hostname].local after installation. Key parameters are:

Parameter	Typical Value	Notes
callsign	WB2MNF	Your FCC callsign — must be exact
grid	FN20	4-character Maidenhead locator for your QTH
power	23	dBm — 23 dBm = 200 mW for TAPR HAT
frequency	10.1387	MHz — 30m dial frequency for TAPR HAT/Pi
tx_percent	20	Transmit 20% of slots (1 in 5)
ppm	0	Crystal frequency correction in PPM
ntp_server	pool.ntp.org	NTP time server — leave default
upload	wspr.live	Spot upload destination

After saving configuration, the system transmits automatically at the next even UTC minute + 1 second. No reboot is required. The WsprryPi systemd service starts automatically on boot and resumes transmitting after any power cycle.

To verify operation: go to wspr.rocks within 10–15 minutes of your first transmission, click the SQL tab, and query:

SELECT * FROM wspr.rx WHERE tx_sign=’WB2MNF’ ORDER BY time DESC LIMIT 10

If your beacon is on the air and being received, spots will appear here within 2–3 minutes of transmission.

10. Quick Reference Summary

Parameter	Value
Protocol type	Continuous-phase 4-FSK
Information content	Callsign + 4-char grid + power (dBm)
Source bits	50 bits (28 callsign + 15 grid + 7 power)
FEC	Rate-1/2 convolutional, K=32
Channel symbols	162 symbols of 2 bits each
Symbol duration	0.6827 seconds (8192/12000)
Tone spacing	1.4648 Hz (12000/8192)
Occupied bandwidth	~6 Hz
Transmission duration	110.6 seconds
Timeslot period	2 minutes (120 seconds)
TX start time	1 second after even UTC minute
Typical duty cycle	20% (one TX per 5 slots)
Min decodable SNR	−28 dB (in 2,500 Hz BW)
Frequency accuracy	±a few Hz (NTP + PPM correction)
30m TX center freq	10.140200 MHz (dial: 10.1387 MHz)
TAPR HAT output	100–200 mW (23 dBm nominal)
Spot database	wspr.live (ClickHouse SQL, 4B+ records)
Live map	wspr.rocks
Software (Pi)	WsprryPi v2.x — github.com/lbussy/WsprryPi

Document prepared by Claude (WB2MNF session) for the GCARC WSPR Network · W2MMD · w2mmd.org · March 2026