How to Troubleshoot Noisy VBT Velocity Readings

A systematic review by Weakley et al. (2021) found that coefficient of variation (CV) for barbell velocity across commercially available VBT devices ranged from 1.1% to 9.7% depending on the device, exercise, and setup conditions — a nearly tenfold difference in measurement noise between best-case and worst-case scenarios. When your device sits at the noisy end of that range, programming decisions based on velocity data become unreliable: load-velocity profiles drift session to session, autoregulation cues fire at the wrong moments, and fatigue monitoring loses its signal in measurement error. Noisy velocity readings are not an inevitable property of VBT technology; they are almost always the product of identifiable, fixable setup and usage errors. This guide walks you through every major cause of measurement noise, a structured diagnostic checklist, and best-practice protocols to restore clean, trustworthy mean concentric velocity data.

Barbell velocity measurement is more sensitive to environmental and technique variables than most coaches expect. All VBT devices — whether inertial measurement units (IMU), linear position transducers (LPT), or optical motion-capture systems — derive velocity from a physical measurement of displacement or acceleration over time. Any factor that adds spurious displacement or acceleration to the signal, or that shifts the sensor's measurement axis relative to the bar's true movement path, generates noise in the output.

The practical consequence is that a reading labeled "1.05 m/s" may contain anywhere from 0.01 to 0.08 m/s of pure measurement error depending on your setup. For load-velocity profiling, that error band can shift a predicted 1RM by 5–15 kg. For within-session autoregulation using a 5% velocity loss threshold, the noise floor may be larger than the threshold itself — making the autoregulation cue meaningless. Fixing the setup is not optional; it is the foundation of every downstream VBT application.

Understanding the mechanism behind each noise source is what makes troubleshooting efficient. Rather than swapping variables at random, a practitioner who knows the physics can isolate the culprit in two to three diagnostic reps.

1. Sensor Placement and Orientation. Every VBT device has a defined measurement axis. An IMU mounted at an angle to the bar's longitudinal axis measures a projection of velocity rather than true vertical velocity. A 10° tilt introduces a cosine error of approximately 1.5%, which is small — but compounded with other errors, it pushes total noise into the clinically meaningful range. For LPT tethers, horizontal displacement of the attachment point from directly below the spool adds a trigonometric error that grows nonlinearly as the bar travels through range of motion.

2. Loose or Shifting Attachment. A sensor that moves independently of the bar — even by a few millimetres per rep — introduces high-frequency acceleration spikes that look like genuine bar movement to the firmware. Velcro straps that loosen across a long session, magnetic mounts on knurled sleeves, and clip-on brackets that rotate slightly between reps are the most common culprits. The noise signature is typically a within-set velocity increase that is not matched by RPE, suggesting the sensor, not the athlete, is moving faster.

3. Bar Whip and Oscillation. At loads above roughly 70% 1RM on a flexible barbell, the bar oscillates vertically at a frequency of approximately 2–5 Hz throughout the lift. A sensor mounted at the collar — the point of maximum oscillation amplitude — measures both true bar velocity and oscillatory velocity superimposed. Courel-Ibáñez et al. (2019) demonstrated that bar whip produced velocity overestimates of 0.04–0.09 m/s on back squats with standard 20 kg barbells, with larger errors at heavier loads and higher rep counts.

4. Range-of-Motion Inconsistency. Mean concentric velocity is averaged across the entire upward phase of the lift. If squat depth varies by 5–8 cm between reps — common under fatigue — the mean velocity calculation integrates over different distances, producing velocity differences that are kinematic artifacts rather than true power-output changes. Consistent ROM is a prerequisite for valid MCV comparison across reps, sets, and sessions.

5. Device Sampling Rate and Filtering. Devices that sample at 50–100 Hz miss the peak velocity portion of fast lifts (jump squats, Olympic derivatives), where the true peak may last only 20–40 ms. Aggressive low-pass filtering set at a cut-off frequency that is too low will smooth out real velocity variation along with noise — creating artificially consistent readings that underrepresent both peak and mean velocity. Pérez-Castilla et al. (2019) compared devices at 50 Hz vs. 1,000 Hz and found mean velocity underestimation of 0.06–0.12 m/s at lighter loads where peak velocity was highest.

6. Bumper Plate Bounce. On Olympic platforms, bumper plates that contact the floor during the eccentric phase can transmit a mechanical impulse through the bar. If the device firmware does not correctly identify the start of the concentric phase after this impact, it may include the post-bounce rebound acceleration in the velocity calculation, inflating MCV by 0.05–0.15 m/s on exercises like deficit deadlifts or touch-and-go bench press.

7. Mixed Lifting Tempo on the Eccentric. Some VBT firmware detects the concentric phase using a velocity threshold (e.g., upward movement above 0.05 m/s). An athlete who pauses at the bottom of a squat for 2–3 seconds before initiating the concentric produces a clean signal; an athlete who transitions smoothly with a stretch-shortening cycle (high eccentric velocity immediately preceding the concentric) may trigger the concentric detection algorithm 50–100 ms early, including a portion of the eccentric deceleration in the mean velocity window and artificially inflating MCV by 0.03–0.07 m/s.

8. Magnetic Interference and IMU Drift. IMUs that incorporate magnetometers for orientation reference can experience drift or spike errors when used near large steel structures (power rack uprights, cable columns, metal flooring), strong magnet systems (certain speaker arrays, proximity to MRI equipment), or when moved rapidly between orientations between sets without allowing the sensor to re-calibrate. This manifests as slow session-long drift in reported velocity (e.g., consistent 0.02 m/s upward trend across a 90-minute session) or single-rep spikes that are physically implausible.

Work through this checklist in order. Each step eliminates a category of noise source. Once velocity stabilizes and reads plausibly for the exercise and load, stop — you have found the root cause.

1. Check the attachment. Remove the sensor and reattach with firm pressure. Run three unloaded (empty bar) reps. Unloaded back squat MCV should be 1.0–1.3 m/s; unloaded bench press 1.0–1.2 m/s. If readings are outside these ranges on an unloaded bar, the device or attachment is the issue, not the exercise variables.
2. Verify axis alignment. Confirm that the sensor's primary measurement axis is parallel to the bar's direction of travel for the exercise. For vertical exercises (squat, deadlift, bench press), the sensor face should be level when the bar is at mid-range. For incline exercises, adjust accordingly. Take a slow video and compare the sensor angle to the bar's trajectory.
3. Standardize ROM. Use a physical depth marker (box or safety bar pin) for squats, a consistent chest touch point for bench press, and a locked knee angle for deadlifts. Measure the range of motion with a tape and log it. Re-run three reps at the same load and compare CV — it should drop by at least 50% if ROM inconsistency was the driver.
4. Switch to a stiffer bar. If available, substitute a 29 mm diameter stiff deadlift bar or a powerlifting bar for the standard 28 mm training bar at loads above 75% 1RM. If readings stabilize, bar oscillation was the driver. If not, continue to the next step.
5. Check for magnetic interference. Move the entire setup 2–3 metres away from rack uprights or other large steel objects. If using an IMU, allow 30 seconds for magnetometer re-initialization. Compare velocity readings before and after relocation.
6. Audit the tempo. Standardize the eccentric tempo across all athletes to a 2-second controlled descent followed by a distinct 1-second pause at the bottom before the concentric. This eliminates SSC tempo as a variable and gives the firmware a clean phase-detection trigger.
7. Review firmware settings. Confirm the device is set to the correct exercise mode. Many devices apply different filtering profiles per exercise — a squat profile may average over 85% of the concentric phase while a jump squat profile integrates through full extension. Applying the wrong exercise mode can shift MCV by 0.10–0.20 m/s.
8. Run a known-load sanity check. Load the bar to exactly 60 kg and perform five back squat reps with a consistent tempo. Reference values: recreational lifters typically produce 60 kg squat MCV of 0.80–1.00 m/s; trained athletes 0.95–1.15 m/s; highly trained 1.10–1.30 m/s. If your reading is outside the plausible range by more than 0.10 m/s, a device-level calibration issue or firmware error is likely.

Use this table as a quick-reference guide when a specific noise pattern appears mid-session. Match the symptom to the most likely cause and apply the fix before collecting more data.

Troubleshooting eliminates noise sources, but validation confirms that the cleaned signal is accurate. Use these three methods to establish confidence in your device's output before committing to velocity-based programming decisions.

Test-Retest Reliability Check. On the same athlete, in the same session, perform two identical sets of three reps at the same load (e.g., 60 kg squat) separated by exactly 5 minutes rest. Calculate the CV between the two sets. A CV below 2.5% indicates acceptable within-session reliability for most VBT applications. A CV above 4% suggests a remaining noise source has not been eliminated. Weakley et al. (2021) reported intraclass correlation coefficients (ICC) above 0.93 for well-controlled LPT setups and 0.87–0.91 for IMU devices under standard training conditions.

Known-Load Sanity Check. Every VBT practitioner should maintain a reference dataset of expected MCV values for key exercises at anchor loads (e.g., 60 kg squat, 80 kg squat, 60 kg bench press). When readings deviate by more than 0.08 m/s from historical averages on a day when the athlete reports normal readiness, a setup error is more likely than a genuine performance shift. Pérez-Castilla et al. (2019) validated a reference MCV table for back squat and bench press across a range of trained athletes that can serve as population-level sanity benchmarks.

Cross-Validation Between Devices. If two devices of different types are available (e.g., an IMU and an LPT), running them simultaneously on the same bar for a set of five reps at a moderate load provides a direct agreement check. Agreement within 0.05 m/s MCV indicates both devices are measuring reliably. Discrepancies above 0.10 m/s indicate one device has an unresolved setup issue. Courel-Ibáñez et al. (2019) demonstrated that IMU and LPT devices converge to within 0.03 m/s MCV under controlled lab conditions, providing a target for field setups.

Each device technology has a characteristic noise profile. Understanding these differences helps practitioners choose the right device for their setting and anticipate which noise sources are most likely to affect their system.

IMU Sensors (Inertial Measurement Units). IMUs use accelerometers and gyroscopes (and sometimes magnetometers) to derive velocity by integrating acceleration over time. Integration errors accumulate with each repetition — a phenomenon called gyroscope drift — making IMUs most accurate for short, single-rep efforts and less reliable over long, multi-rep sets without a firmware re-zeroing algorithm. IMUs are the most portable and lowest-cost option, with no cable to snag, but they are most susceptible to magnetic interference and mounting angle errors. Best suited for: jump squats, Olympic lift derivatives, single-rep testing.

Linear Position Transducers (LPT / String Potentiometers). LPTs measure velocity by tracking the rate at which a tethered cable extends from a spool as the bar moves upward. Because velocity is derived directly from displacement (not double-integrated from acceleration), LPTs are generally more accurate than IMUs over multi-rep sets and are considered the field-standard reference device in most validation research. Their primary noise sources are cable angle error (when the spool is not directly below the bar), cable snag, and bumper-plate bounce transmitting through the tether. Best suited for: squats, deadlifts, bench press, all exercises with consistent vertical bar path.

Optical Systems (High-Speed Camera or LiDAR). Optical systems track reflective markers or the bar itself using high-frame-rate cameras (typically 240–1,000 fps) or laser distance measurement. They offer the highest accuracy and are immune to magnetic interference and cable-angle errors, but they require controlled lighting, fixed camera positions, and substantial post-processing. In field settings, environmental factors (changing ambient light, bar shadow, bar rotation) introduce noise that partially offsets the theoretical accuracy advantage. Optical systems are primarily used for lab-based research validation rather than day-to-day strength and conditioning programming.

The practical implication: if your IMU readings are noisy and a reliable LPT is available, switching to the LPT for slow-grind exercises (squats, deadlifts, bench press) at high loads may resolve the issue without any further troubleshooting. Reserve the IMU for ballistic exercises where cable tethers are impractical.

Eliminating noise permanently requires building these practices into the standard session protocol, not applying them reactively after problems emerge.

Before the session: Mount the sensor to the same sleeve position on the same barbell every session. Mark the sleeve with a permanent marker ring to ensure consistent placement. Initialize the device and allow 60 seconds for IMU calibration. Confirm the exercise mode matches the exercise. Log the barbell diameter, sleeve position, and exercise mode in your session notes.

During warm-up: Run three reps at an empty bar before collecting any data. Confirm that empty-bar MCV is in the expected range for the exercise (squat: 1.0–1.3 m/s; deadlift: 1.2–1.6 m/s; bench press: 1.0–1.2 m/s). If readings are outside range, troubleshoot before adding load. Do not proceed to working sets with unresolved noise on the warm-up data.

During working sets: Use a physical ROM marker for every exercise to ensure consistent depth. Call out the target MCV to the athlete before each set — this reduces tempo variation because athletes who know they are being measured converge toward a more consistent movement pattern. Discard any rep flagged by the device as an outlier (most modern firmware has this feature) and do not include it in set averages.

Between sets: Do not move the device between exercises without re-checking attachment. For IMU systems, allow 10–15 seconds of sensor stillness before the next set to enable gyroscope re-zeroing if the firmware supports it. For LPT systems, confirm the cable is fully retracted and not kinked before beginning the next set.

Session to session: Maintain a consistent pre-session calibration protocol. If using multiple devices, assign each device a fixed athlete ID so that session-to-session drift in a specific device does not contaminate the group dataset. Re-run the known-load sanity check (three reps at the reference load) at the start of every third session to detect slow drift before it affects programming decisions.

Noisy velocity readings are a setup problem, not a technology problem. The following six practices eliminate the vast majority of VBT measurement noise encountered in real training environments:

1. Standardize sensor placement. Same sleeve position, same axis alignment, same attachment torque — every session, every exercise.
2. Standardize ROM. Use physical markers for depth-dependent exercises. A 5 cm ROM variation produces more velocity noise than most sensor calibration errors.
3. Control tempo. Enforce a consistent eccentric tempo and bottom-pause protocol so the firmware's concentric detection algorithm triggers at the same point every rep.
4. Match device to exercise. Use LPT for heavy vertical exercises; use IMU for ballistic and Olympic derivatives where cable tethers are impractical. Apply the correct firmware exercise mode.
5. Run sanity checks regularly. An empty-bar check at session start and a known-load reference set every third session catch setup errors and device drift before they corrupt programming decisions.
6. Validate with test-retest. A within-session CV below 2.5% at a reference load confirms the setup is clean. Above 4% means a noise source remains. Use the symptom table in this guide to isolate it efficiently.

With these practices in place, mean concentric velocity becomes a precise, reproducible measurement that genuinely reflects the athlete's neuromuscular output on each rep — which is the entire premise of velocity-based training.