added internal references to paper

Author: zeyus

packages/liblsl/paper/paper.md

@@ -62,17 +62,17 @@ Performance was characterized under controlled conditions to isolate different s
 
 ## Results and Discussion
 
-![Latency characterization of the liblsl Dart package. **(A)** Local performance on iPad Pro M4 (left) and Pixel 7a (right), each producing and consuming its own 1000 Hz data stream. **(B)** Network performance between two iPads over 1Gbps Ethernet. Violin plots show median (solid line) with first and third quartiles (dashed lines). See Methods for experimental details and Table 1 for complete statistics. *Note: Outliers >500µs were excluded from visualization but are included in statistics*.](./figures/plot_latency.png)
+![\label{latencyfig} Latency characterization of the liblsl Dart package. **(A)** Local performance on iPad Pro M4 (left) and Pixel 7a (right), each producing and consuming its own 1000 Hz data stream. **(B)** Network performance between two iPads over 1Gbps Ethernet. Violin plots show median (solid line) with first and third quartiles (dashed lines). See Methods for experimental details and \autoref{latencystats} for complete statistics. *Note: Outliers >500µs were excluded from visualization but are included in statistics*.](./figures/plot_latency.png)
 
 | Condition | Device      |      n | Min (µs) | Max (µs) | Mean (µs) | SD (µs) |
 |:----------|:------------|-------:|---------:|---------:|----------:|--------:|
 | Local     | iPad Pro M4 | 180000 |       21 |     4264 |        65 |      47 |
 | Local     | Pixel 7a    | 179900 |       42 |    18809 |       274 |     434 |
 | Network   | iPad Pro M4 | 180000 |       60 |     4261 |       148 |     129 |
 
-Table: Summary of latency measurements.
+Table: \label{latencystats} Summary of latency measurements.
 
-Table 1 summarizes latency statistics across conditions and devices. Figure 1 illustrates the distribution of the relevant performance characteristics: (A) single-device performance showing the API's computational overhead when producing and consuming data locally (iPad: µ=65µs, σ=47µs; Pixel 7a: µ=274µs, σ=434µs), where standard deviations reflect timing jitter inherent to the operating system's thread scheduling and (B) network performance between two iPads over a 1Gbps wired connection (µ=148µs, σ=129µs). These results confirm that the Dart wrapper introduces minimal overhead beyond the base LSL C library performance, with observed differences primarily attributable to device hardware capabilities and network infrastructure rather than the API implementation itself.
+\autoref{latencystats} summarizes latency statistics across conditions and devices. \autoref{latencyfig} illustrates the distribution of the relevant performance characteristics: (A) single-device performance showing the API's computational overhead when producing and consuming data locally (iPad: µ=65µs, σ=47µs; Pixel 7a: µ=274µs, σ=434µs), where standard deviations reflect timing jitter inherent to the operating system's thread scheduling and (B) network performance between two iPads over a 1Gbps wired connection (µ=148µs, σ=129µs). These results confirm that the Dart wrapper introduces minimal overhead beyond the base LSL C library performance, with observed differences primarily attributable to device hardware capabilities and network infrastructure rather than the API implementation itself.
 
 The observed latencies are consistent with previous benchmarking of native LSL implementations [@kotheLabStreamingLayer2025] and demonstrate that this package preserves the real-time performance characteristics required for neurophysiological applications [@iwamaTwoCommonIssues2024]. The local processing latencies (65-274µs) are well below typical neurophysiological event timing requirements (<1ms), confirming suitability for EEG, electromyography (EMG), and other biosignal applications[^1]. The mean inter-device network latency (148µs) demonstrates that network overhead remains minimal on well-configured local networks[^2], though researchers should note that wireless networks and other factors including network traffic congestion may introduce additional jitter. The device-dependent variation (iPad: 65µs vs Pixel 7a: 274µs) reflects differences in CPU architecture, device network hardware and operating system scheduling rather than API limitations, indicating that platform selection should be validated through careful testing based on the requirements of the specific application.