Game Boy Advance Audio Interpolation: Cleaning Up Classic Game Sound

The Game Boy Advance's audio hardware, while innovative for its time, produced sound that many modern players find harsh and noisy. A new blog post from jsgroth explores how emulators can dramatically improve this audio experience through advanced interpolation techniques, transforming the way classic GBA games sound.

The core problem stems from the GBA's audio architecture. The hardware outputs mixed audio samples using pulse-width modulation (PWM) at one of four possible sampling frequencies ranging from 32,768 Hz to 262,144 Hz, with most games using 65,536 Hz. However, the GBA resamples from each audio channel's frequency to the PWM sampling frequency using simple nearest-neighbor interpolation - essentially just outputting the channel's current sample. This creates extremely noticeable audio aliasing, particularly when games use very low sample rates with the two PCM channels, which is common in many titles.

The Solution: Enhanced Interpolation

The approach described in the blog post is conceptually simple: instead of accurately emulating the GBA's PWM hardware, emulators can use their own interpolation algorithms to resample from audio channels' sample rates directly to the emulator's output sample rate (typically 48,000 Hz). This breaks from accurate emulation but produces significantly cleaner audio.

The first step involves calculating each PCM channel's source sample rate using the clock divider and counter reload value of the GBA timer that the channel is tracking. Games generally use consistent sample rates across most of their audio, though some titles like Castlevania: Circle of the Moon switch between different rates for different songs.

Interpolation Algorithms Compared

The blog demonstrates two main interpolation approaches:

Cubic Hermite Interpolation - A 6-point algorithm that works well for upsampling and provides good quality with reasonable performance. The implementation uses a specific formula that balances smoothness with computational efficiency.

Windowed Sinc Interpolation - A more complex, band-limited approach that produces extremely high-quality results by eliminating audio aliasing more completely. However, this comes at the cost of significantly more computational complexity and can sometimes sound overly muffled, especially with lower sample rates.

Real-World Examples

Several comparison examples illustrate the dramatic differences:

From Metroid: Zero Mission (13,379 Hz sample rate):

• Accurate resampling: Harsh, aliased audio
• Cubic Hermite: Much cleaner, slightly muffled
• Windowed sinc: Least aliased but noticeably muffled

From Mega Man Zero 4 (21,024 Hz sample rate):

• All three versions show improvement, with windowed sinc best at reducing static/hissing
• The higher sample rate makes the differences less pronounced

From Golden Sun: The Lost Age (63,072 Hz sample rate):

• Even at this high rate, sinc interpolation removes some audio noise
• Cubic interpolation provides good improvement but doesn't eliminate noise as effectively

The PSG Challenge

A significant complication arises with the PSG (Programmable Sound Generator) channels. When games use both PCM and PSG channels together, the PSG output still contains high-frequency wave components that stand out against the cleaner PCM audio. This creates an imbalance where the PSG channels sound too prominent.

The solution involves applying a low-pass filter to the PSG output with a cutoff frequency set to half the PCM sample rate. This approach works well in most cases but can create issues when the PSG leads the music while PCM channels play secondary instruments.

Implementation Considerations

The blog notes that this approach works well for cleaning up GBA audio in real-time, though the performance impact can be significant, primarily due to the PSG downsampling requirements. The implementation dynamically sets the low-pass cutoff frequency based on the current PCM sample rate, designing a new 2nd-order Butterworth filter each time the sample rate changes.

While this enhanced interpolation cannot completely eliminate audio noise (since it still receives noisy 8-bit samples as input), it significantly reduces audio aliasing and removes some noise, providing a much more pleasant listening experience for classic GBA games.

This technique represents a practical compromise between accurate emulation and audio quality, similar to how many emulators implement graphics enhancements that deviate from hardware accuracy to provide better visual experiences. For players who prioritize audio quality over perfect hardware accuracy, these interpolation techniques offer a compelling way to breathe new life into classic GBA soundtracks.