RESEARCH — SYSTEM ALGORITHM
ZLC
Zero-Latency Convolution
ZLC applies arbitrarily long FIR filters with no added latency: a direct path handles the first samples while partitioned FFT blocks process the tail. The workhorse under every DAM Audio product — reverbs, room simulation, feedback suppression.
THE PROBLEM
Convolution always had a price
Block-based FFT convolution adds its block size in latency; direct time-domain FIR melts the CPU as impulse responses grow. Real-time systems were forced to choose between late and heavy.
HOW IT WORKS
How it works
Partitioned scheme
Gardner-style partitioning: a direct FIR path plus exponentially growing FFT partitions.
Truly zero added latency
The first sample out depends only on the first sample in — safe for tracking, live and monitoring.
RT-safe threading
A worker pool spreads partitions across cores with no allocation and no priority inversion on the audio thread. IR updates are click-free crossfades.
UNDER THE HOOD
Going deeper
ZLC partitions the work Gardner-style: the first 64–128 samples run as direct time-domain FIR — zero latency by construction — while exponentially growing FFT partitions process the tail, however long it is.
A worker pool spreads partitions across cores with no allocation and no priority inversion on the audio thread, and IR updates crossfade equal-power in about 10 ms: filters can change live without a click. The same engine powers our reverbs, room simulation and feedback suppression.
AT A GLANCE
At a glance
| Input | Any FIR filter set |
| Output | Convolved audio, unlimited IR length |
| Latency | 0 ms added |
| Implementation | C++17 · worker pool · RT-safe |
| Availability | Ships in every DAM Audio product · OEM licensing |
PARAMETERS
Hands on the algorithm
| Direct path | 64 – 128 samples, pure time domain |
| Partitions | Exponentially growing FFT blocks |
| IR length | Unlimited |
| IR updates | Click-free equal-power crossfade |
| Threading | Worker pool, RT-safe, no allocation on audio thread |
POSITIONING
Compared to the alternatives
vs block FFT convolution
Block processing adds its block size in latency; ZLC's direct path makes it zero.
vs direct FIR
Pure time-domain convolution at these IR lengths would melt the CPU; partitioning keeps it real time.
APPLICATIONS
Where it fits
Audio plugins
Convolution reverbs and IR-based processors usable while tracking.
Live processors
Long correction filters in latency-critical signal chains.
Embedded DSP
Deterministic CPU profile suits fixed-resource platforms.
Monitoring & simulation
Room simulation per monitor position with 0 ms penalty.
INTEGRATION
Built to live inside your product
| Delivery | C · C++ · MATLAB · .dsp — full source code |
| Platforms | macOS · Windows · Linux · embedded ARM |
| DSP platforms | Flow DSP · Audio Weaver — solutions in preparation |
| Documentation | Whitepaper — every algorithm explained, in the clear |
OEM LICENSING
- ■ One-time payment per brand
- ■ Full source code — C, C++, MATLAB, .dsp
- ■ Whitepaper — all algorithms explained
- ■ Integration support included
- ■ Free updates
- ■ Volume discounts on multiple licenses
SHIPS IN