When Your Tennis Serve Speed Gun and IMU Disagree: Ball Speed vs. Racket Speed Explained

A junior academy in Florida had this same argument three times last season before anyone bothered to write down what was actually being compared. The radar gun behind the baseline clocked a flat serve at 187 km/h. The racket sensor clipped under the kid's overgrip that same session reported 118 km/h. His dad pulled both apps up side by side, decided one of them was broken, and spent twenty minutes recalibrating a sensor that had never stopped working correctly.

Neither number was wrong. They were never measuring the same object, and no amount of recalibrating a working device closes a gap that physics put there on purpose. A speed gun clocks the ball after it has already left the strings - an aerodynamic projectile in free flight. An IMU clipped to a racket, wrist, or forearm clocks the swinging implement itself, at a different point in the collision, and often at a different point on the athlete's body than the one that actually contacts the ball. Coaches who don't separate these two signals end up either mistrusting a device doing exactly what it's built to do, or coaching an athlete toward a racket-speed target that was never physically attainable at that measurement point.

What follows is what each device actually reads, why the ball number is almost always the larger one, what two independent research groups found measuring both sides of that gap, and a protocol for running radar and IMU together so the difference becomes useful data instead of a parking-lot argument.

Doppler radar measures the velocity of whatever is currently reflecting its signal back at it. Once the ball separates from the strings, that's the ball - a compressed rubber sphere already in flight, with no further energy being added to it. Everything the racket, arm, and legs did up to that point is baked into a single number at the instant of separation, and the radar has no way to unpack it.

An IMU works from the opposite end of the chain: a package of accelerometers and gyroscopes fixed to a physical object, reporting that object's own linear and angular velocity, nothing downstream of it. Where that package sits changes what number comes out. A unit near the racket throat reports something close to true racket-head speed. A unit built into a wrist strap or forearm sleeve reports limb speed instead, meaningfully lower, because the wrist is the pivot driving a lever - the racket head - that travels a longer arc at higher tangential velocity than the joint producing it. Baseball pitching coaches hit the identical problem comparing a radar-clocked pitch to an arm-sleeve IMU's reported arm speed; the two numbers describe different links in one kinetic chain, not competing measurements of the same thing.

Neither device is more accurate than the other here. They were never measuring the same physical quantity, so a mismatch between them isn't evidence that either one drifted out of calibration.

At impact, a nearly stationary tennis ball collides with a much heavier, fast-moving racket. Basic collision physics says the lighter object comes off that collision moving faster than the heavier one was moving going in - the same reason a bat sends a ball off faster than the bat itself was swinging, or a golf clubhead's speed is lower than the ball speed it produces. The string bed adds a second boost on top of that mass mismatch: modern strings behave like a trampoline, storing and returning energy through the collision rather than absorbing it, which pushes the ball's exit speed further above the racket's speed than a simple mass ratio alone would predict.

The combined effect is an amplification factor, not a rounding error. Across the serve biomechanics literature, ball exit speed typically runs somewhere between 1.4 and 1.7 times the racket-head speed measured at the moment of contact, with the exact multiplier shifting with string tension, string pattern, ball condition, and how close to the racket's sweet spot the strike lands. A racket-mounted IMU reading 30 m/s on a serve whose ball speed reads 48 m/s off a radar isn't disagreeing with the radar. It's reporting the input half of a collision whose output the radar happens to be measuring.

Fleisig, Nicholls, Elliott and Escamilla (2003) measured serve mechanics in a sample of high-level male servers using radar positioned directly behind the baseline, in line with serve trajectory, on every trial - the controlled placement needed for a trustworthy ball-speed number at all. Their servers produced ball velocities of roughly 51 to 63 m/s (184-227 km/h). The study's own limitation is worth sitting with: a small, elite sample under lab-grade radar placement tells you what's physically possible at the top end, not what a club-level setup measuring a recreational server should expect, and the ball-speed figure alone says nothing about how fast that server's racket head was moving to produce it.

That second half of the picture is where wearable-sensor research fills the gap. Whiteside, Cant, Connolly and Reid (2017), working with Tennis Australia's biomechanics group, mounted inertial sensors on players' rackets and used machine-learning models to classify shot type and estimate hitting load from the raw motion data, validating output against synchronized video. Their models classified shot types with roughly 80-90% accuracy depending on category, but the paper is explicit about a limitation that matters here: the sensor reports the racket's own motion signature, and turning that signature into a precise, ball-speed-equivalent number required model-specific calibration. It was never a drop-in substitute for a radar reading, and the authors didn't treat it as one. Racket-mounted IMU data is a genuinely useful, independent signal about swing mechanics and load - it was never designed to reproduce a ball-flight number, and expecting it to is where most club-level confusion actually starts.

Put four measurement tools on the same serve and you'll get four defensible numbers, because each one captures a different physical point in the same event. Values below are approximate elite flat-serve figures; recreational servers see the same pattern at lower absolute numbers.

Notice that the two IMU rows alone differ by roughly a factor of two, purely from mount position - before radar even enters the comparison. Mixing wrist-sensor data with throat-mounted sensor data in the same trend line is a more common source of session-to-session confusion at most clubs than the radar-versus-IMU gap that gets all the attention.

Most disagreements that end with someone reaching for a screwdriver trace back to one of these five habits, not a hardware fault.

• Treating a lower IMU number as a malfunction. A racket sensor reading roughly half the radar number on a clean flat serve is doing its job - that's the expected amplification ratio, not an error code.
• Comparing sensors mounted at different points as interchangeable. A wrist-strap reading from one session and a throat-mounted reading from another never belong on the same trend line, regardless of brand.
• Setting an IMU alert threshold from a radar-derived number. A cue built around hitting 51 m/s on the racket sensor targets a number that measurement point can't physically produce.
• Ignoring the ratio as a diagnostic. A flat serve well under a 1.4 ratio - racket speed steady, ball speed dropping - usually points to mistimed or off-center contact, not weak effort.
• Assuming the ratio holds across serve types. Kick and topspin serves convert some racket energy into ball rotation instead of linear speed, lowering the ratio for reasons unrelated to swing strength.

A session that logs both devices produces more useful data than either one alone, provided the setup stays disciplined about what's being compared.

1. Fix and log the IMU mount point before starting - throat, handle, or wrist - and never blend two mount points into one athlete's trend line.
2. Set the radar to standard placement: behind the baseline, centered on the service box, within roughly 5 degrees of the intended serve line.
3. Record 6-10 serves with both devices running at once, discarding any trial with a mistimed radar trigger or an audible off-center strike.
4. Calculate the ball-to-racket ratio per clean trial, not just the two raw numbers - the ratio shows whether contact quality changed, not just power.
5. Track the ratio over weeks alongside the raw numbers. A steady racket speed with a quietly dropping ratio flags a timing issue earlier than waiting for ball speed to fall.
6. Re-baseline after any equipment change - new racket, string tension, or a moved sensor mount - since each shifts the ratio independent of the athlete's own output.

These ranges are compiled from serve biomechanics literature as a working band, not a strict cutoff - string setup, ball condition, and IMU mount position can shift raw numbers even when the underlying physics hasn't changed.

The kick serve's lower ratio isn't a red flag by itself - it's the signature of energy being deliberately redirected into spin rather than linear ball speed. What's worth flagging is a flat serve sitting well under 1.4, or any serve type's ratio dropping noticeably from an athlete's own established baseline while racket speed stays flat. That pattern points at contact quality - a slightly mistimed release or an off-center strike - well before it shows up as a slower ball off the radar.