Case studies

Fifteen measured results, from a 15 fs clock for AI optics and a 59 GHz oscilloscope in a data-center lab to a bracelet that runs on body heat. Each one gives the setup, the instrument and the source document. Click any chart to enlarge it.

NewAI data center and optics

A 625 MHz clock at 15.73 fs, lab-grade from its own supply

15.73 fsRMS jitter at 625 MHz, 12 kHz to 20 MHz

A 625 MHz ultra-low-jitter LVDS differential oscillator for optical modules, AI servers, SerDes and high-speed networking, with Harmony on its 3.3 V supply.

With Harmony, RMS jitter measured 15.73 fs from 12 kHz to 20 MHz on the 625.09 MHz carrier. Integrated phase noise is −87.2 dBc.
Without a filter, supply spurs stand out of the phase-noise curve from about 10 kHz to 20 MHz, the largest near 200 kHz. With Harmony, every one of them is gone.
The floor holds at −161 dBc/Hz from 2.9 MHz to 20 MHz, the performance you would expect from a lab-grade supply.

Setup. Tai-SAW 625 MHz LVDS differential oscillator at 3.3 V, without a filter and with Harmony on its supply. Agilent (Keysight) E5052B signal source analyzer, correlation 4. Tai-SAW Technology is a TransSiP PI solutions partner: small quantities through Tai-SAW, 1,000 pieces or more from TransSiP.

Source: TransSiP 625 MHz demo, Agilent E5052B measurement

Ask about the 625 MHz clock Where to buy it

Agilent E5052B phase-noise plot of a 625 MHz carrier: the green trace without a filter shows supply spurs between 10 kHz and 20 MHz, the yellow trace with Harmony is smooth, with 15.73 fs RMS jitter from 12 kHz to 20 MHz

Power conversion

ADI’s LT8622S Silent Switcher, made noise-free

2,232×lower output ripple

Harmony and two 47 µF capacitors at the 1 V, 2 A output of the LT8622S evaluation board.

Ripple fell from 10 mVp-p to 4.48 µVp-p. The 1.97 MHz fundamental now reads −103.3 dBm.
Mean noise density from 10 kHz to 100 MHz fell from 25.1 to 10.7 nV/√Hz.
Switching harmonics and microwave spurs were removed.
Harmony revealed the Silent Switcher’s true noise floor: 20 dB lower than the ADI datasheet shows.

Setup. ADI LT8622S Gen 3 Silent Switcher evaluation board, with Harmony Type-B at VOUT. The related video series was made in collaboration with ADI and Keysight.

Source: ADI LT8622S Noise Free Silent Switcher poster (2025)

Read the measurement Watch the video

Poster: LT8622S evaluation schematic with a Harmony PI Filter at VOUT, before and after noise-density plots, and a spectrum marker at −103.30 dBm showing 2,232 times lower ripple

NewUltra-low-noise LDOs

ADI’s LT3046, made battery-like and noise-free

−130 dBmnoise floor at 3.3 V, with every spur removed

The newest result, shared by Nokia Bell Labs: two LT3046 ultra-low-noise LDOs in series, fed from a noisy switcher.

A noisy SMPS at 6 V feeds a first LT3046 at 5 V, then a second LT3046 at 3.3 V.
Without Harmony, spurious noise still reached about −100 dBm, even after two of ADI’s ultra-low-noise LDOs.
With Harmony Type-B after the second LDO, every spur was removed. The 3.3 V rail sits on the −130 dBm (10 nV/√Hz) noise floor from 1 Hz to 6 GHz.

Setup. Noisy SMPS, then LT3046 at 5 V, then LT3046 at 3.3 V, with and without a Harmony Type-B evaluation board. Swept from 1 Hz to 6 GHz at 50 Hz resolution bandwidth.

Source: Courtesy of Nokia Bell Labs, Integrated Circuits & Microsystems Research Lab

Harmony with LDOs

Nokia Bell Labs bench: a noisy SMPS feeding two ADI LT3046 LDOs and a Harmony Type-B board, and a 1 Hz to 6 GHz spectrum where spurs at about −100 dBm without Harmony disappear with Harmony, leaving a −130 dBm floor

AI data center and optics

A flat switching rail inside a high-speed laser source

0.95 V, 4 Awith no visible ripple

Validated by a market leader in a high-speed laser source, on a 59 GHz real-time oscilloscope.

Without Harmony, the loaded 0.95 V switching rail shows millivolts of ripple at 2 mV/div.
With Harmony, the trace is flat on the 100 µV scale.
It is the first flat, ripple-free SMPS bias we have seen at this bandwidth.

Setup. Keysight UXR0594A real-time oscilloscope: 59 GHz, 256 GSa/s, 10-bit. SMPS at 0.95 V and 4 A load, August 2025.

Source: Noise Free Innovations 26V2, p. 8

Applications in the data center

Keysight oscilloscope screen: a 0.95 V rail without Harmony shows ripple, while the rail with Harmony is a flat line near 100 µV

Clocks and timing

Clock phase noise with 200 mVp-p on the supply

18–30 dBlower phase noise, 20 kHz to 2 MHz

A differential crystal oscillator for data-center links, tested with deliberate supply noise.

At 20 kHz offset, phase noise improved from −123.8 to −153.3 dBc/Hz.
At 1 MHz offset, it improved from −144.1 to −162.4 dBc/Hz.
The improvement narrows to 1.4 dB at 10 MHz, where the oscillator’s own floor dominates.

Setup. Power-supply phase-noise (PSPN) test: 200 mVp-p signal injected through a bias tee, XO alone versus XO with Harmony.

Source: Noise Free Innovations 26V2, p. 9

See the full curve

Test setup: a Mini-Circuits bias tee feeding a Harmony board ahead of a differential XO board

Wireless and 5G

Lab-grade 5G clock jitter from a noisy 3.3 V rail

233 → 20 fsRMS jitter, 12 kHz to 20 MHz

A 122.88 MHz VCXO powered from a noisy 3.3 V PDN, the way it runs in real 5G equipment.

On the noisy PDN alone, RMS jitter measured 233 fs.
With Harmony Type-B on the same rail, it measured 20 fs. The lab reference with a Keysight E5052B clean bias is 18 fs.
An earlier, unoptimized demo board measured 32 fs.

Setup. Keysight E5052B signal source analyzer. Harmony Type-B EVB connected by SMA between the supply and the VCXO VCC.

Source: TransSiP Noise Free ULPN 122.88 MHz VCXO 25V1; Quick Start Guide 25V3, §9

Read the analysis

A VCXO test board connected by SMA cable to a Harmony Type-B evaluation board

Ultra-low-noise LDOs

TI’s lowest-noise LDO, taken to the analyzer floor

3 Hz–6 GHzat the spectrum analyzer’s noise floor

Measured with Nokia Bell Labs: a TPS7A94 fed from a noisy 3.9 V DC-DC.

With the TPS7A94 alone, noise above 100 kHz was not suppressed, despite the filters under the EVB’s metal lid.
With Harmony Type-B at the LDO’s 3.3 V output, the noise reached the floor of the Keysight PXA analyzer.
Harmony Type-B on the noisy 3.9 V DC-DC alone, with no LDO, gave the same result.

Setup. TI TPS7A94 EVB converting 3.9 V to 3.3 V, swept from 3 Hz to 6 GHz on a Keysight PXA signal analyzer.

Source: Harmony Quick Start Guide 25V3, §6

Harmony with LDOs

Two spectrum analyzer screens, without and with Harmony on a TI TPS7A94 output, above a photo of the noisy 3.9 V switcher board and the TPS7A94 EVB

Defense radio

An ADM7150’s baseband noise, reduced to power-off levels

20 dBless 1/f noise

An operational test by a defense entity on an ultra-low-noise, high-PSRR LDO inside a radio system.

1/f noise fell 20 dB from 1 Hz to 100 Hz.
Noise fell 10 dB from 1 kHz to 100 kHz, and 20 dB from 200 kHz to 100 MHz.
No high-frequency ground bounce remained. The active LDO measured like an LDO switched off.

Setup. ADM7150 in a radio system under real-world conditions, with Harmony and a bulk capacitor at its output. Reproduced by permission.

Source: Noise Free Innovations 26V2, p. 13; EVB Read Me, p. 14

Harmony with LDOs

Noise spectrum from 1 Hz to 100 MHz: LDO active without Harmony, LDO off, and LDO active with Harmony at power-off levels

PDN noise

Three parts against a 15-part 6 GHz filter

93%less PCB area, and a noise-free rail

A high-noise 3.3 V rail from a Picotest VRTS3 board, filtered two ways.

TI’s 6 GHz broadband filter on the LP877451Q1EVM uses 15 passives. It left the rail “still noisy, just enough to pass EMI”.
Harmony Type-B uses three parts. It brought the rail to −99.97 dBm, the analyzer’s own floor, from 9 kHz to 3.2 GHz.
The filter area shrank from 112 mm² to 7.7 mm².

Setup. Picotest VRTS3: a TI LMR10515 buck converting USB 5 V to 3.3 V, with its 125 MHz and 10 MHz clocks on. Rigol DSA832 spectrum analyzer.

Source: Why Classic Target Impedance Approach Not Effective 25V3; Noise Free Innovations 26V2, p. 5

Why target impedance falls short

A Picotest VRTS3 board feeding a red Harmony Type-B evaluation board through SMA adapters into a coaxial cable

PDN impedance

A PDN that is flat in micro-ohms

7,400×lower peak impedance

Shunt impedance of a working PDN from 100 kHz to 3 GHz, before and after Harmony.

Before: resonances of 1.13 Ω at 127.8 MHz and 584 mΩ at 532 MHz.
After: 116 to 339 µΩ from 1 MHz to 3 GHz.
With no resonances, there is no phase change across the band.

Setup. Vector network analyzer, 2-port shunt-through impedance, 100 kHz to 3 GHz.

Source: Noise Free Innovations 26V2, p. 11

See the full curve

Impedance plot before Harmony with sharp resonance peaks up to about 1.1 ohms — Before

Impedance plot after Harmony, flat near zero across 100 kHz to 3 GHz — After

RF front ends

Two Skyworks front-end modules, 16 decoupling parts to 3

16 → 3decoupling components

A mission-critical radio powered from a noisy PMIC.

The original design used LC filters and capacitors from 1 pF to 4.7 µF around each FEM. They produced high-Q resonance peaks.
One Harmony and two 47 µF capacitors now feed VCC_1, VCC_2 and VSUP1 on both modules.
Both FEMs operate at the noise floor, with no filters around the chips.

Setup. Two Skyworks RF FEMs on a customer board. Reproduced with customer permission.

Source: Noise Free Innovations 26V2, p. 15

Why bypass networks top out

Customer PCB with two RF front-end modules and no decoupling parts around them, a Harmony filter, and the PMIC marked as noisy

IoT and research

NYU WIRELESS: a faster, more reliable BLE link

12.5 dBless supply noise at the receiver

Published at IEEE WiSNeT 2025 by researchers at NYU Tandon’s RAISE Lab.

The LDO output ripple fell from 59 mV to 14 mV, a 12.5 dB reduction.
The receiver chain gained at least 10 dB of SNR, and the bitrate doubled at 50 m.
The point where SNR collapses moved more than 30 m further out.

Setup. An nRF24L01 2.4 GHz link on Arduino boards, with a Harmony filter in series with the LDO and receiver VDD for A/B testing.

Source: D. Wong, A. Padmanathan and H. Rahmani, “Performance Optimization of Resource-Constrained Wireless Receiver Front-Ends with Power-Centric Design Approaches,” IEEE WiSNeT 2025, pp. 46–49

Paper on IEEE Xplore

12.5 dB

less supply noise at the receiver

GNSS

Navigation that stays on the trail

10×better positioning accuracy

TransSiP PI inside GNSS sports watches, bike computers and a dual-band module.

Two watches with the same GNSS chipset ran a 5 km forest trail. One recorded 5.27 km without TransSiP PI, and the other recorded 5.03 km with it.
On a route repeated 10 times, including a two-story covered garage, the PI-equipped bike computer’s tracks overlapped consistently.
In a drive test, off-road excursions at bends, curve overshoot and erroneous trajectories all dropped with Harmony.

Setup. GNSS products using TransSiP PI. They include a 16 × 12 mm dual-band L1/L5 module built on Sony’s CXD5610, which runs rails as low as 0.65 V and 0.50 V.

Source: Noise Free Innovations 26V2, pp. 16 to 19

10× smaller GNSS position errors

Aerial view of a route ridden ten times, with tightly overlapping tracks, including inside a covered garage

Frontier research

Supply-noise elimination at 4 K for quantum computing

DC–20 GHzat 4 K (about −269 °C)

Collaborative research with industry experts on cryogenic power delivery.

The goal is power-supply noise elimination from DC to 20 GHz at cryogenic temperature.
This builds on the dual Type-B configuration, which has been measured to 20 GHz at room temperature.

Setup. Research collaboration. Details are shared under agreement.

Source: Noise Free Innovations 26V2, p. 10

Talk to us about cryogenic use

DC–20 GHz

at 4 K (about −269 °C)

Biomedical wearables

A body-heat-powered health tracker in Japan

4TransSiP PI products inside

MEDIROM’s MOTHER smart bracelet uses TransSiP PI noise-reduction technology. It runs on body heat and needs no charging.

Optical heart-rate accuracy with a single LED and detector, at minimal power.
In cycling, limits of agreement against a chest strap were +11 / −12 bpm. Published multi-LED designs measure +13 / −27 bpm.

Setup. Single-LED optical heart-rate monitor with Harmony on the LED driver and the AFE, detector and current-ADC rails.

Source: Noise Free Innovations 26V2, pp. 20 to 22. Multi-LED reference: BMC Sports Science, Medicine and Rehabilitation 10:10 (2018).

Biomedical applications

TransSiP PI products inside

Try it on your rail.

The Type-B evaluation board connects between your supply and your load with two SMA cables.

Get the evaluation board Buy TSNJ-5A6L26