Gold Code

A Gold code is generated by the modulo-2 addition (XOR) of two carefully selected maximal-length sequences (m-sequences), known as a preferred pair. These parent sequences must have the same length $N = 2^n - 1$ and exhibit a specific three-valued cross-correlation function. By cyclically shifting one of the m-sequences and XORing it with the other, a family of $2^n + 1$ distinct sequences is produced, including the two original m-sequences. This construction method ensures that the resulting set is larger than what a single m-sequence can provide while maintaining strict correlation guarantees.

The defining feature of a Gold code family is its three-valued cross-correlation function. For any two sequences in the set, the cross-correlation takes only three possible values: $-1$, $-t(n)$, or $t(n)-2$, where $t(n)$ is a function of the sequence length. This tight bound is significantly lower than the worst-case cross-correlation between arbitrary sequences of the same length. This property minimizes multiple access interference (MAI) in CDMA systems, allowing multiple users to transmit simultaneously on the same frequency without catastrophic mutual disruption.

Within a Gold code family, sequences are categorized as balanced or unbalanced based on the number of binary ones and zeros in one full period. A balanced Gold code has exactly one more '1' than '0', which suppresses the DC component of the carrier and provides optimal spectral properties. Unbalanced codes have a disparity greater than one, leading to residual carrier leakage. For practical spread spectrum systems, balanced Gold codes are preferred because they ensure better carrier suppression and more uniform power spectral density.

Unlike a single m-sequence, which has a perfect two-valued autocorrelation (peak at zero shift, constant low value elsewhere), a Gold code exhibits a four-valued autocorrelation function. The peak at zero shift is $N$, but the off-peak values can be $-1$, $-t(n)$, or $t(n)-2$. This slightly inferior autocorrelation compared to pure m-sequences is the trade-off for obtaining a much larger family of sequences with excellent cross-correlation. The off-peak autocorrelation sidelobes are still bounded and predictable, which is critical for initial code acquisition and synchronization.

The most famous deployment of Gold codes is the Coarse/Acquisition (C/A) code in the Global Positioning System (GPS). Each GPS satellite broadcasts on the same L1 frequency (1575.42 MHz) using a unique 1023-chip Gold code from a family generated by two 10-bit linear feedback shift registers. The bounded cross-correlation allows a terrestrial receiver to distinguish and track up to 32 satellites simultaneously using code division multiple access. The C/A code's 1.023 Mcps chip rate provides a sequence period of 1 millisecond.

The existence of a Gold code family depends on identifying a preferred pair of primitive polynomials for the linear feedback shift registers (LFSRs). Two polynomials of degree $n$ form a preferred pair if their corresponding m-sequences have a cross-correlation that takes only the three specific values. Not all primitive polynomials have a preferred pair. For example, for $n=5$, the polynomials $x^5 + x^2 + 1$ and $x^5 + x^4 + x^3 + x^2 + 1$ form a valid pair. These polynomial pairs are tabulated in literature for standard register lengths.

The Coarse/Acquisition (C/A) code transmitted on the GPS L1 frequency (1575.42 MHz) is a Gold code of length 1023 chips, generated by combining two 10-stage linear feedback shift registers. Each satellite is assigned a unique code phase, enabling a receiver to distinguish signals from up to 32 space vehicles simultaneously through code division multiple access. The low cross-correlation ensures that the weak signal from one satellite is not masked by a stronger signal from another, which is critical for accurate pseudorange measurement and position triangulation.

In the IS-95 and cdma2000 cellular standards, Gold codes serve as long code masks and pilot sequences to separate base stations and users on the reverse link. Unlike synchronous Walsh codes, Gold codes maintain low cross-correlation even without global time alignment, making them ideal for the asynchronous mobile environment. The family of codes generated from preferred pairs of m-sequences provides a large pool of distinct addresses, allowing network operators to reuse codes across geographically separated cells without catastrophic interference.

3G UMTS networks utilize hierarchical Gold codes for cell separation on the downlink. The primary scrambling code is a complex-valued sequence formed from the modulo-2 sum of two real m-sequences. The standard defines 512 primary scrambling codes, divided into 64 groups of 8, to facilitate the three-step cell search procedure:

• Step 1: Slot synchronization using the primary synchronization code (PSC)
• Step 2: Frame synchronization and code group identification using the secondary synchronization code (SSC)
• Step 3: Primary scrambling code identification through symbol-by-symbol correlation over the CPICH

Gold codes are employed in spread spectrum transponders for secure distance measurement and low-probability-of-intercept telemetry. The large family size allows unique code assignment to individual assets, while the noise-like spectral properties hide the signal beneath the thermal noise floor. In deep-space communications, Gold codes provide the ranging signal for precise orbit determination. The receiver performs a code phase search by correlating the incoming signal with a local replica, and the measured offset directly yields the round-trip light time.

The balanced spectral properties and large code set of Gold sequences make them suitable for spread spectrum watermarking in digital media. A unique Gold code is embedded into the content at low amplitude, acting as a robust fingerprint that survives compression and format conversion. The low cross-correlation ensures that multiple watermarks can coexist without mutual destruction, enabling collusion-resistant traitor tracing. The watermark detector correlates the suspect content against all assigned codes; a peak above the decision threshold identifies the source of the leak.

IEEE 802.11b Direct Sequence Spread Spectrum (DSSS) physical layers use a Barker sequence for the preamble, but the high-rate DSSS (HR/DSSS) mode employs complementary code keying (CCK) derived from code families with Gold-like correlation properties. In research prototypes for code-divided sensor networks, Gold codes enable multiple low-power nodes to transmit simultaneously to a central sink without packet collisions. The sink employs a bank of parallel correlators, one for each sensor's assigned code, to demultiplex the overlapping transmissions in real time.