The noise on the noise
Switching supplies are efficient and noisy, while linear regulators are quiet and wasteful. We found a third problem underneath both: switching noise that moves in time. Harmony is built to remove it.
The problem with DC-DC conversion
Noise-sensitive systems need a constant supply voltage, free of disturbances. Switch-mode supplies give the best efficiency and waste the least battery or harvested energy, but they produce a very noisy supply bias. Linear regulation gives a cleaner rail at the expense of efficiency, and that cost grows as on-chip operating voltages fall.
Even with linear regulation, simultaneous switching noise (ground bounce) rises with I/O activity, and conventional filters cannot keep up with it. Power-saving duty cycles add transients on top. Each generation answers with more complex power management and more filter parts, and the noise remains.
Switching Noise Jitter
Switching Noise Jitter (SNJ) is a time-domain form of noise that sits on top of the familiar frequency-domain picture of a switching supply: its power spectral density, ripple and harmonics. SNJ is switching noise whose phase moves in time. It is the dominant noise that degrades signal integrity under weak signal conditions.
We discovered SNJ while investigating instability in GPS systems. Conventional frequency-domain analysis could not see it, so we turned to real-time spectral analysis. Tektronix has published a case study on how we used its RSA306 spectrum analyzer with DPX to capture these short, time-varying events.
How Harmony removes it
Harmony is an ultra-wideband, non-linear passive device. Its patented SNJ conditioning works in the time domain as well as the frequency domain. It enables DC power delivery at the thermal or system noise floor of the DC-DC, LDO or PDN behind it.
Its attenuation gets wider and deeper when it is cascaded with bulk capacitors. In the Type-B configuration, two 47 µF capacitors at VOUT create an anti-resonance near 1 MHz with S21 close to the noise floor. Harmony then neutralizes the capacitors’ inductance from that point upward, so the floor holds out to multiple gigahertz.
This is why one circuit works on every rail. Classic PDN design tunes a capacitor ladder to a target impedance measured with 50 Ω instruments (VNA and TDR), but a real PDN is not a 50 Ω system. Harmony’s result does not depend on the source or load impedance.
A real PDN, before and after
Marker readings from a VNA shunt-through measurement, 100 kHz to 3 GHz. The 1.13 Ω resonance at 127.8 MHz falls to 153 µΩ, about 7,400 times lower.
Shunt impedance of a real PDN, before and after Harmony
- Before: capacitor network
- After: Harmony Type-B
Source: Noise Free Innovations 26V2, p. 11 (PDN case study)
Harmony against a classic target-impedance network
| Property | Capacitor, ferrite bead and LDO network | Harmony PI Filter |
|---|---|---|
| Noise suppression | About 80 to 90 dB at best, in a limited band | About −120 dB with a single Type-B, and −140 to −160 dB with dual Type-B. Both are below the noise floor of most instruments: measured traces sit at the VNA floor (about −100 dB) from 2 MHz to 500 MHz. |
| Effective bandwidth | Fades out near 300 to 400 MHz | DC to 6 GHz in Type-B. 62 dB at 10 GHz and 51 dB at 20 GHz in dual Type-B. |
| PDN impedance | A rising target profile, for example 65 to 170 mΩ | 104 to 255 µΩ in Type-B, and 2 to 5 µΩ in dual Type-B from 14 MHz to 300 MHz |
| Resonance | Anti-resonance peaks, LC tank ringing and overshoot | No resonance and no phase change across the band |
| Parts for a 6 GHz filter | 15 (TI LP877451Q1EVM broadband filter) | 3 (one Harmony and two 47 µF capacitors) |
| Board area | 112 mm² | 7.7 mm², 93% less |
| Design portability | Each PDN is tuned to its own load and is not portable | One circuit for every rail, independent of source and load impedance |
| Time-domain noise (SNJ) | Not addressed | Conditioned in both the time and frequency domains |
| Power drawn | An LDO post-regulator costs dropout headroom | None. The part is passive. |
Sources: datasheet R009; Quick Start Guide 25V3; “Why Classic Target Impedance Approach Not Effective” 25V3. For the industry background, see “The Myth of Three Capacitor Values” by Eric Bogatin, Larry Smith and Steve Sandler.
Patents
Issued in the United States and China.
| US 9,748,663 | Metamaterial substrate for circuit design |
|---|---|
| US 10,319,685 | EMI-shielded integrated circuit packages and methods of making the same |
| US 10,942,219 | Time and frequency domain signal conditioning device for switching noise jitter (SNJ) reduction, and methods of making the same |
| US 11,243,249 | Circuit arrangement for switching noise jitter (SNJ) reduction in feedback control loop circuits, and methods of making the same |
| ZL 2017 8 0008665.9 | EMI-shielded integrated circuit packages and methods of making the same (China) |
| ZL 2017 8 0018642.6 | Time and frequency domain signal conditioning device, circuit arrangement, and methods of making the same (China) |