Resolution analysis in a lens-free on-chip digital holographic microscope

1. Introduction

• Conventional whole-slide imaging requires mechanical scanning and stitching to combine high resolution with a large FOV, increasing complexity and cost.Computational microscopy approaches, including lens-free on-chip microscopy, offer high-resolution wide-FOV imaging without those steps.Source: §1 · p. 1
• Lens-free on-chip microscopy provides effective NA near 1 across the sensor’s native wide FOV without intermediate optics.Its unit-magnification configuration places samples close to the sensor; LFOCDHM reconstructs images from diffraction patterns formed by interference.Source: §1 · p. 2
• LFOCDHM resolution remains far below the ideal coherent diffraction limit despite pixel super-resolution, and five system factors have not been rigorously examined together.The factors are sample-to-sensor distance, illumination’s spatial and temporal coherence, sensor-pixel size, and reconstructed sub-FOV extent.Source: §1 · p. 2
• The paper derives five transfer-function models, combines them into a unified model, and uses simulations and experiments to assess their validity and design utility.The models account for the five factors and their interactions, and are intended to predict resolution limits or guide parameter selection.Source: §1 · pp. 2–3

2.1. Typical optical setup for LFOCDHM

• A source illuminates the specimen, and scattered and transmitted light interfere at the sensor; diffraction patterns encode object amplitude and phase for computational reconstruction.The sample is typically 0.3–2 mm from the detector because protective glass covers the sensor.Source: §2.1 · p. 3
• The resolution model separates defocus, temporal and spatial coherence, finite pixel size, and reconstructed sub-FOV extent into transfer functions whose least cut-off frequency limits resolution.The four coherence, pixel-size, and reconstructed-region transfer functions are mutually independent; propagation yields absorption and phase transfer functions.Source: §2.1 · p. 4

2.2. Theoretical analysis of resolution in LFOCDHM

• Defocus trades improved low-frequency phase response at larger distances against more transfer-function zero-crossings; multi-height measurements reduce these zeros but leave very-low-frequency phase recovery difficult.For practical distances of several hundred microns, increasing distance can improve reconstruction somewhat; the work uses single-height measurements thereafter.Source: §2.2.1 · pp. 5–6, Figure 3 · p. 6
• Wider spectral width lowers the temporal-coherence cut-off frequency, while narrower bandwidth improves resolution; diffraction-limited performance requires z2 < 2λ^2/∆λ, although phase recovery needs adequate defocus.When sensor constraints prevent sufficiently small z2, the passage recommends a narrower-band source; it notes sensor defocus cannot go below 300 µm.Source: §2.2.2 · pp. 7, 9
• Larger illumination-source diameter reduces spatial-coherence cut-off frequency; smaller sources or a larger source-to-sample distance improve spatial coherence and resolution.For a 200 µm source and 0.5 µm ideal half-pitch limit, the passage gives z2/z1 < 1/200; with z2 > 400 µm, z1 must exceed 80 mm.Source: §2.2.3 · pp. 9–10
• Finite pixel size is modeled through binning and uniform-interval sampling; binning removes spectral information beyond its cut-off, and sampling mixes high-frequency information into the low-frequency domain.The stated finest reconstructed half-pitch feature is ∆p/w for pixel pitch ∆p and w ≥ 1.Source: §2.2.4 · pp. 10–12

• As down-sampling factor w increases, pixel aliasing creates more zero-valued interlaced frequency regions, while recovering w-fold higher resolution requires w^2 raw images.At w = 2, PSTF(u) tends to zero only at the spectrum periphery; for w > 2, additional interlaced regions lose spectral information.Source: §2.2.4 · pp. 12–13

2.2.5. Influence of the finite extent of reconstructed sub-FOV on imaging resolution

• A finite reconstructed sub-FOV excludes some high-angle diffraction patterns, reducing resolution; at ΔL = 61µm, half-pitch resolution falls to 2µm.For a 400µm sample-to-sensor distance and NA ∼0.8, the sub-FOV side length must be at least 2845µm; resolution also worsens toward its borders, creating a tradeoff between retaining high frequencies and practical computation.Source: §2.2.5 · pp. 13, 15

2.2.6. Comprehensive influence of multiple factors on imaging resolution

• For the described conventional system, pixel size limits resolution without pixel SR, whereas source spectral width principally constrains reconstruction when pixel SR is used.The paper identifies spectral width as another main obstacle to resolution improvement that numerical methods alone are difficult to solve or alleviate.Source: §2.2.6 · p. 15

2.2.7. Optimization of the imaging resolution for a LFOCDHM system

• The models guide resolution-oriented choices: improve source coherence, use smaller pixels, minimize sample-to-sensor distance, and reconstruct the largest practical sub-FOV centered on the target.A small pinhole or narrow band-pass filter can improve spatial or temporal coherence, respectively; for phase objects, multi-height retrieval with larger sample-to-sensor distances supports reliable phase recovery, especially at low frequencies.Source: §2.2.7 · pp. 15–16

3.1. Experimental setup

• Experiments used a broadband source, CMOS sensor, and controlled changes to illumination coherence, pixel size, and reconstructed-region size.The setup included selectable central wavelengths, bandpass filters, and pinholes.Source: §3.1 · pp. 16–17, Figure 10 · p. 16

3.2. Influence of temporal coherence on imaging resolution

• With spectral widths of 20nm and 30nm, measured resolutions were 1.953µm and 2.461µm, versus theoretical values of 1.936µm and 2.371µm, respectively.The two bandwidths were tested using bandpass filters.Source: §3.2 · p. 17, Figure 11 · p. 17

3.3. Influence of spatial coherence on imaging resolution

• At ∆s = 1.0 mm and z1 = 4 cm, measured resolution was ∼6.20 µm versus theoretical half-pitch 5.81 µm; at fixed z1, smaller ∆s provided higher resolution.For ∆s = 1.3 mm, experimental results agreed well with theoretical values.Source: §3.3 · p. 18

3.4. Influence of pixel size on imaging resolution

• For sensor pixel sizes of 1.67µm, 2.2µm, 3.75µm, and 4.4µm, measured resolution agreed with theoretical values limited by pixel size.The source-to-sample distance was made large enough to exclude spatial-coherence effects.Source: §3.4 · pp. 18–19, Figure 13 · p. 19

3.5. Influence of the reconstructed region on imaging resolution

• Reducing the reconstructed region from a 198µm to a 110µm side length lowered resolution from at least 1.74µm to 3.10µm.The measured object’s location within the sub-FOV also mattered: a group-8 target element was distinguishable in one region but not a nearby one; only the central region reliably achieved expected resolution.Source: §3.5 · pp. 19–20, Figure 14 · p. 20

4. Conclusions and Discussions

• Sensor pixel size is identified as the most limiting factor; smaller pixels can improve super-resolution quality and reduce the information required to reach a target resolution.Other factors are described as relatively easy to handle, while larger pixels require higher up-sampling and produce frequency gaps that pixel super-resolution cannot recover.Source: §4 · p. 20
• Increasing source-to-sample distance and decreasing sample-to-sensor distance can improve coherence and imaging resolution; decreasing sample-to-sensor distance can also reduce the reconstructed area for a fixed target resolution.Using an LED offers a more compact, portable, and low-cost system, but its coherence length affects reconstructed resolution.Source: §4 · pp. 20–21
• Transfer-function cutoffs quantify resolution limits, and experiments confirmed the validity of the theoretical models.The minimum first cutoff frequency determines the quantitative limit, enabling preliminary estimates of ultimate resolution after super-resolution methods.Source: §4 · p. 21
• Computational compensation for imperfect system parameters is proposed as future work, using forward image formation models and inverse-problem algorithms.The paper demonstrates resolution improvement through optical-design optimization and says post-processing compensation will be addressed in future work.Source: §4 · p. 21