Quantum Computing RF

Cosine Pulse

Q: How does a cosine pulse differ from a Gaussian pulse for qubit gates?

Both shapes ramp the drive envelope smoothly to limit spectral leakage, but they differ in their tails and bandwidth. A cosine (raised-cosine, or Hann) pulse reaches exactly zero amplitude at its endpoints over a finite duration with no truncation, so it has clean compact support and is trivial to generate on an arbitrary waveform generator. A Gaussian pulse has infinite theoretical extent and must be truncated, typically at 2 to 3 standard deviations, which leaves a small amplitude discontinuity at the edges. The cosine pulse has a slightly wider main spectral lobe than a tightly truncated Gaussian but lower far-out sidebands because of its continuous derivative at the endpoints. In practice both are paired with a DRAG correction on the quadrature channel to suppress leakage into the qubit's second excited state.

Q: What gate duration does a cosine pulse need on a transmon qubit?

For a single-qubit X or X/2 rotation, the cosine pulse duration is set by the available drive power and the leakage budget rather than by the cosine shape itself. On a typical transmon with an anharmonicity of -200 to -300 MHz, gate durations of 15 to 40 ns are common. A shorter pulse needs higher peak amplitude to deliver the same integrated rotation angle (the area under the envelope), and its wider bandwidth excites more population into the |2> state. A cosine envelope of 20 ns places its first spectral null at about plus or minus 100 MHz from the carrier (roughly a 200 MHz null-to-null main lobe), so on a qubit with 250 MHz anharmonicity the leakage from shaping alone stays below about 0.1 percent before DRAG correction. Designers trade gate speed against coherence time T2 and leakage to find the sweet spot.

Q: Why is a cosine pulse preferred over a square pulse for qubit control?

A square (rectangular) pulse turns the drive on and off instantaneously, producing a sinc-shaped spectrum whose sidebands fall off only as 1/f and extend far beyond the carrier. Those sidebands overlap the qubit's neighboring transitions and the readout resonator, driving unwanted population transfer and leakage. A cosine pulse ramps the amplitude continuously, so its spectrum rolls off much faster (the Hann window's sidebands fall as 1/f cubed, more than 31 dB below the main lobe), confining the energy near the carrier frequency. This keeps the gate selective to the intended 0-to-1 transition and dramatically reduces leakage into the second excited state, which is the dominant coherent error source in fast superconducting-qubit gates.

/KOH-syn puhls/

Used to drive single-qubit gates, this microwave control envelope ramps the drive amplitude up and down along a half-cosine (raised-cosine, or Hann) shape rather than switching on abruptly. The smooth turn-on and turn-off confine the pulse energy near the carrier so that little spectral content spills onto neighboring transitions, which is the chief mechanism by which a fast gate on a qubit leaks population into the second excited state. Because the cosine envelope reaches exactly zero at both ends over a finite span, it is simple to synthesize on an arbitrary waveform generator and is often combined with a DRAG quadrature correction to push leakage below 0.01%. The peak amplitude is scaled so the integrated area sets the desired Rabi rotation angle, typically π for an X gate.

Category: Quantum Computing RF

Typical Duration: 15 to 40 ns

Sideband Roll-off: 1/f³ (> 31 dB)

Shaping the Qubit Drive Envelope

In a superconducting quantum processor, single-qubit logic gates are performed by sending a short microwave burst at the qubit transition frequency (commonly 4 to 8 GHz) down the control line. The carrier rotates the qubit on the Bloch sphere, but it is the envelope (the slowly varying amplitude that multiplies the carrier) that determines the spectral footprint of the gate. A square envelope switches the drive on and off instantaneously and therefore radiates a sinc-shaped spectrum with broad, slowly decaying sidebands. Those sidebands overlap the |1⟩-to-|2⟩ transition, which sits only one anharmonicity away (roughly 200 to 300 MHz below the qubit frequency on a transmon), and they pump unwanted population into that leakage state.

The cosine pulse replaces the abrupt edges with a continuous raised-cosine ramp. Its amplitude starts at zero, rises along the first half of a cosine, holds or peaks at center, and returns to zero, with a continuous first derivative at the endpoints. This is mathematically the Hann window familiar from spectral analysis, and it carries the Hann window's favorable property: far-out sidebands that decay as 1/f cubed instead of the square pulse's 1/f. The result is a gate whose energy is tightly clustered near the carrier, sharply reducing the bandwidth that overlaps adjacent transitions.

The trade-off is bandwidth versus speed. A shorter gate needs a wider envelope spectrum, which raises leakage; a longer gate is spectrally clean but exposes the qubit to more decoherence over the gate window. Designers tune the pulse duration against the qubit anharmonicity and coherence time, then layer a DRAG correction on the quadrature channel to cancel the residual leakage that the cosine shape alone cannot remove.

Envelope and Rotation Equations

Raised-Cosine (Hann) Envelope:
Ω(t) = (A / 2) × [1 − cos(2πt / t_g)] for 0 ≤ t ≤ t_g

Rotation Angle (pulse area):
θ = ∫₀^t_g Ω(t) dt = A × t_g / 2

Peak Amplitude for a π Gate:
A_π = 2π / t_g (so θ = π)

First Spectral Null (bandwidth):
Δf_null ≈ 2 / t_g from the carrier; null-to-null width ≈ 4 / t_g
(e.g. t_g = 20 ns → first nulls at ±100 MHz, ≈ 200 MHz main-lobe width)

Where Ω(t) = drive Rabi envelope, A = peak Rabi amplitude, t_g = gate duration, θ = Bloch-sphere rotation angle. A 20 ns cosine gate keeps shaping-only leakage near 0.1% on a transmon with ≈250 MHz anharmonicity.

Control-Pulse Envelope Comparison

Envelope	Sideband Roll-off	Endpoint Value	Compact Support	Leakage (pre-DRAG)	Typical Use
Square	1/f (slow)	Step discontinuity	Yes	1 to 5%	Calibration, idle
Cosine (Hann)	1/f³	Exactly zero	Yes	~0.1%	Single-qubit gates
Gaussian (truncated)	Near-Gaussian	Small step at edge	No (truncated)	0.1 to 0.5%	Single-qubit gates
Cosine + DRAG	1/f³ + I/Q	Exactly zero	Yes	< 0.01%	Fast high-fidelity gates
Flat-top cosine	1/f³ edges	Exactly zero	Yes	Length-dependent	Two-qubit, parametric

Common Questions

Frequently Asked Questions

How does a cosine pulse differ from a Gaussian pulse for qubit gates?

Both shapes ramp the envelope smoothly to limit leakage, but a cosine (Hann) pulse reaches exactly zero at its endpoints over a finite span, giving clean compact support that is trivial to generate on an AWG. A Gaussian has infinite theoretical extent and must be truncated at 2 to 3 sigma, leaving a small edge discontinuity. The cosine has a slightly wider main lobe but lower far-out sidebands (1/f³) thanks to its continuous derivative. Both are typically paired with DRAG to suppress |2⟩-state leakage.

What gate duration does a cosine pulse need on a transmon qubit?

Duration is set by available drive power and the leakage budget, not by the cosine shape itself. On a transmon with −200 to −300 MHz anharmonicity, 15 to 40 ns is typical. A 20 ns cosine envelope places its first spectral null at about ±100 MHz from the carrier (roughly a 200 MHz null-to-null main lobe), so on a 250 MHz anharmonicity qubit shaping-only leakage stays near 0.1% before DRAG. Shorter pulses need higher peak amplitude and wider bandwidth, increasing leakage; the designer trades speed against T2 and leakage.

Why is a cosine pulse preferred over a square pulse for qubit control?

A square pulse switches the drive instantaneously and radiates a sinc spectrum whose sidebands fall only as 1/f, spilling energy onto neighboring transitions and the readout resonator. The cosine envelope ramps continuously, so its sidebands fall as 1/f³ (more than 31 dB below the main lobe), confining energy near the carrier. That keeps the gate selective to the 0-to-1 transition and sharply cuts leakage into the second excited state, the dominant coherent error in fast superconducting-qubit gates.

Related Terms

← CoSaMP Cosine Taper →