High-Resolution Speckle-based Single-Pixel Imaging using Silicon Photonic Non-Redundant Optical Phased Array

arXiv:2608.20922 · physics.optics · Submitted 2026-08-21 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "High-Resolution Speckle-based Single-Pixel Imaging using Silicon Photonic Non-Redundant Optical Phased Array".

Mira: Here's the gist of the paper. Optical phased arrays (OPAs) are promising wavefront controlling devices for imaging applications due to their compact and high-speed nature Here, they experimentally demonstrate high-resolution imaging using a non-redundant OPA (NR-OPA) Due to the non-redundant antenna layout of NR-OPA based on the Costas array, the resolution scales quadratically with N Combined with the speckle-based single-pixel imaging (SSPI) scheme, which avoids the need for precise phase calibration, they experimentally achieve a large number of resolvable imaging points exceeding 10,000 using a silicon photonic NR-OPA chip with only N = 127 without sweeping the wavelength The demonstrated scheme provides a promising route toward mega-pixel imaging by a compact OPA chip with reduced number of phase shifters.

Kai: Concept of Speckle-based Single-Pixel Imaging The SSPI system involves splitting the input light into N waveguides within the OPA chip Light passing through each waveguide is phase controlled by a phase shifter and emitted from a grating antenna By driving N phase shifters of the OPA using random electrical signals, speckle-like far field patterns (FFPs) are generated by interference of the light from N emitters on the OPA.

Mira: The generated intensity patterns are used to illuminate the imaging target, and the total optical power transmitted from the target is detected by a single-pixel photodetector (PD).

Kai: To retrieve the image, illumination patterns are switched for K times, and K PD signals are recorded.

Title and authors: Kai: So we're looking at this paper now, "High-Resolution Speckle-based Single-Pixel Imaging using Silicon Photonic Non-Redundant Optical Phased Array," and the authors are a team from The University of Tokyo and ETH Zurich. They’re talking about how they built this thing.

Mira: Yeah, they're focusing on using a silicon photonic non-redundant optical phased array, or NR-OPA, for imaging applications because it’s compact and fast. It sounds like they are tackling the resolution limits of conventional arrays by changing the antenna layout.

Lev: From an engineering standpoint, I want to know what they actually built. Are we talking about a lab setup or something that can run on real hardware? And what kind of fidelity are we talking about with these silicon components?

Kai: They used a silicon-on-insulator wafer with a two hundred twenty nanometer thick Silicon layer and a two micrometer thick buried oxide layer for the chip. The actual OPA setup included an edge coupler for the fiber input, and then a cascade of one by two multi-mode interference couplers to split the light into one hundred twenty eight waveguides, with one port just for monitoring.

Mira: And they controlled those waveguides using one hundred twenty seven thermo-optic phase shifters, which is what you need to drive the system. They placed those emitters on a Costas array, which is a specific type of antenna arrangement that they use for this non-redundant setup.

Lev: So if I try to translate that into what would run on real hardware, the key thing is how they implemented those phase shifters—they used two hundred twenty micrometer long titanium nitride heaters with one kiloohm electrical resistance and four micrometer wide trenches for thermal isolation. That sounds like a setup where you'd need pretty precise control over the temperature to get that two pi phase shift.

Kai: Exactly, and they managed to achieve that full two pi phase shift at approximately twenty milliwatts of power using those heaters, which is what drove the system. They also used grating couplers for the emitters on that Costas array, with a grid pitch of fifteen micrometers.

Mira: That fifteen micrometer grid pitch resulted in a field of view of five point nine two degrees when they were using a one five five zero nanometer wavelength. It gives you some context on the spatial coverage they are looking at with this setup.

Title and authors: Lev: five point nine two degrees is pretty tight for an imaging system, but I get it if you're aiming for high resolution over a small area. So, what’s the main takeaway here about how this setup compares to older methods?

Kai: The main point they are making is that conventional periodic OPAs only scale their spatial resolution linearly with the number of antennas N. But because this NR-OPA uses a non-redundant layout based on the Costas array, the resolution scales quadratically with N.

Mira: That quadratic scaling is what’s important there. It means you get a much better resolution boost just by adding more elements, rather than just adding more elements linearly. They are showing that this scaling is a direct consequence of having those mutually distinct displacement vectors between every pair of antenna elements, which they call r prime i j equals r i minus r j (b) (c) (a).

Lev: If the resolution scales quadratically with N, that’s a big win for complexity. But how does that quadratic scaling translate into actual numbers for the number of resolvable points? That's where I need to see the math on this page.

Kai: They evaluate this by looking at something called rank P, which they say indicates the amount of independent information they get about an object o from their equations. And for this non-redundant OPA, because those displacement vectors are mutually distinct, the imaging resolution is maximized to N squared minus N plus one.

Mira: That formula N squared minus N plus one is what sets the theoretical upper bound for the number of resolvable points they can get without doing any wavelength sweeping. They’ve shown that with just N equals one hundred twenty-seven antennas, this gives them a certain number of resolvable points, and they claim this is the largest number they've obtained in an integrated-OPA-based imaging system without wavelength scanning <ref:2608.20922#pg3>.

Lev: One two seven antennas for that result is actually pretty small if you think about building a scalable system for real hardware. So, what’s the actual performance metric on the numbers? How many points can they actually resolve experimentally?

Kai: They achieved a large number of resolvable imaging points exceeding ten thousand by using this silicon photonic NR-OPA chip with only N equals one two seven without having to sweep the wavelength at all. That's quite impressive for an integrated system.

Mira: And they did this by combining it with a speckle based single pixel imaging scheme, which is clever because it gets around the need for precise phase calibration that usually messes up these kinds of systems. They generate speckle-like far field patterns by driving those one hundred twenty seven phase shifters with random electrical signals.

Title and authors: Lev: So the system relies on those random signals to create the illumination pattern, and then a single pixel photodetector detects the resulting intensity transmitted from the target object o. What does that look like in terms of practical implementation?

Kai: To retrieve the image, they switch those illumination patterns for K times and record K photodetector signals. They state that each signal sk can be expressed as p k times o, where pk is the k-th illumination pattern and o is the transmittance distribution of the target object.

Mira: The experimental results show that the autocorrelation of that illumination pattern has a sharp peak at the center with a full width at half maximum of zero point zero seven millimeters, which they equate to zero point zero four degrees. That tells you about how fine-grained their speckle pattern is.

Lev: zero point zero four degrees is quite small for an imaging system, that suggests really high spatial frequency information is being captured, even if the overall field of view is limited by the costas array pitch. What does this mean for feature extraction?

Kai: They confirmed that their signal-to-value ratios for the illumination matrix P remained relatively large up to xi equal to one six zero zero three, which they say is exactly N squared minus N plus one, matching their simulated results quite well.

Mira: And when they reconstructed an image of a target, they got a peak signal-to-noise ratio of thirteen point nine decibels. This is a solid number for the quality of the image reconstruction using this method.

Lev: Thirteen point nine decibels is respectable, but I have to ask about the limitations mentioned. What are they saying about where this system stops working or what it doesn't do?

Kai: They did flag that while they achieved high resolution for line and space patterns with widths as narrow as eighty eight point four micrometers, the total number of resolvable points is large, but they also noted that their setup is limited to a single wavelength.

Mira: So the main limitation seems to be that it's currently fixed to one wavelength, which means they aren't doing the full mega-pixel imaging yet without adding more complexity. They are aiming for a route toward mega-pixel imaging with a compact OPA chip that uses fewer phase shifters.

Title and authors: Lev: So, summarizing this "High-Resolution Speckle-based Single-Pixel Imaging using Silicon Photonic Non-Redundant Optical Phased Array," the core contribution is proving that non-redundant antenna layouts based on Costas arrays can give you quadratic scaling of resolution with N, which beats the linear scaling of uniform arrays.

Kai: Exactly. It’s a way to get better spatial fidelity without having to sweep the wavelength, which is a major hurdle in OPA systems right now.

Mira: And for AI systems, this means they can achieve high-resolution feature extraction at the micro-scale, like detailed object recognition or detecting tiny defects in manufacturing. It simplifies things because they don't need that precise phase calibration step to get started.

Lev: For me on real hardware, the quadratic scaling is exciting conceptually, but building a chip with one hundred twenty seven active phase shifters and all those components is still a lot of overhead if you want to scale up much further than this initial test case.

Kai: That’s fair, the complexity of driving those phase shifters with pseudo-random voltage signals in the range of two volt to five volt, where the phases are proportional to the square of those voltages, is definitely a factor we have to consider for real deployment.

Mira: And since they achieved this using only one hundred twenty seven emitters and without wavelength sweeping, they’re setting a promising route toward mega-pixel imaging with a much smaller number of phase shifters overall.

Lev: So the paper shows the potential of this architecture—the quadratic scaling—but the real work for me is figuring out how to make that scaling practical and robust when you start thinking about thousands of antennas instead of just one hundred twenty seven.

Kai: Right, so they’ve proven the concept works experimentally at this level, showing that we can get ten thousand resolvable points right now using this silicon photonic NR-OPA chip.

Mira: It’s a solid demonstration of how to use speckle based single pixel imaging to overcome some of the traditional calibration headaches in OPA systems.

Lev: So, the main thing to remember about "High-Resolution Speckle-based Single-Pixel Imaging using Silicon Photonic Non-Redundant Optical Phased Array" is that it shows quadratic resolution scaling with N, which is a significant advantage over linear scaling.

Kai: It’s a compact OPA chip that can achieve this high resolution without needing wavelength sweeping for this specific demonstration.

Mira: That sets up a clear path toward achieving mega-pixel imaging with reduced numbers of phase shifters down the line.

The paper's summary: Kai: So we just looked at how they built this silicon photonic non-redundant optical phased array for imaging, and now we’re going to look at their summary and what that really means for high-resolution imaging.

Mira: Right, so the main idea is that they took a regular array—the kind where the antennas are in a grid—and changed it into something non-redundant using this costas array setup.

Kai: Exactly, because of this non-redundant layout, they found that their spatial resolution doesn't just scale linearly with the number of antennas N anymore. It scales quadratically with N.

Mira: That quadratic scaling is what’s really interesting here because it means you can get much finer detail just by adding more elements, instead of having to sweep a wavelength to improve resolution in the first place.

Kai: So they achieved this using only one hundred twenty seven antennas without changing the wavelength at all, and they managed to get over ten thousand resolvable imaging points.

Mira: That’s a big number for an integrated chip because it shows that you can get high spatial frequency information on demand without the complexity of phase calibration.

Kai: It also uses this speckle-based single-pixel imaging scheme, which means they don't need to be super precise with the phase shifters to get a usable image. They just use random electrical signals to create interference patterns.

Mira: So if you’re someone who only listens to the show and want the big picture, this paper shows that we can design compact optical systems where resolution scales quadratically, which is a major step for future AI vision systems needing micro-scale detail.

Kai: The caveat is that they're currently stuck at one specific wavelength, so it’s not quite full mega-pixel imaging yet without more hardware changes.

Mira: True, they are aiming for that mega-pixel goal with fewer phase shifters down the road because this non-redundant architecture cuts down on the total number of shifters needed compared to traditional designs.

Kai: We'll talk about how they actually set up these physical components next and what those numbers tell us about the actual hardware feasibility.

The paper's improvements: Kai: So, we just talked about how they built this non-redundant array for imaging at a single wavelength, and now we’re looking at what improvements they suggest for pushing this further out.

Mira: They are suggesting that while the current setup works well, the next step involves moving away from a single wavelength limitation. They want to explore how you can make this system work across multiple wavelengths without having to sweep them.

Kai: That makes sense because sweeping wavelengths is slow and complex for hardware, so they are looking for a way to keep the high resolution while having flexibility in the light source.

Mira: And they also focus on reducing the number of phase shifters required. They want to see if there’s a way to use fewer active components while still keeping that quadratic scaling benefit we talked about earlier.

Kai: Because right now, driving all those one hundred twenty seven phase shifters with pseudo-random voltage signals in that two volt to five volt range is quite demanding on the control electronics.

Mira: They are looking at ways to optimize the illumination pattern generation itself so that you don't need such a massive number of random inputs to get a good speckle pattern.

Kai: This points toward smarter methods for controlling the phase distribution, maybe something that’s less reliant on brute-force electrical signals and more based on structured light or some other optical trick.

Mira: If they can reduce the shifters while keeping the resolution scaling quadratic, that would be a huge win for making this kind of imaging practical on a real chip.

Kai: It’s about making it compact and efficient enough to move beyond this one hundred twenty seven antenna setup and actually build something that scales up nicely.

Mira: So the focus shifts from just getting the points onto the screen, to making the whole system robust and scalable for larger applications in AI vision pipelines.

Kai: We'll talk about what they’re actually trying to do next with these new ideas, focusing on how they plan to test these improvements in practice.

Conclusion: Kai: So we're wrapping up this look at "High-Resolution Speckle-based Single-Pixel Imaging using Silicon Photonic Non-Redundant Optical Phased Array." We saw how they used a non-redundant antenna layout to get quadratic scaling in resolution, even at only one wavelength.

Mira: That quadratic scaling is the key thing here, because it shows that with just more antennas, you get a much better spatial resolution than if you just added them in a uniform grid.

Kai: Right, and they did this using only one hundred twenty seven emitters without needing to sweep the wavelength at all to get those high-resolution imaging points over ten thousand.

Lev: From my side, it’s impressive that they achieved that level of performance on an integrated silicon photonic chip, but I have to ask how much overhead you’d actually need for a system that scales up past one hundred twenty seven antennas.

Mira: They are really setting a promising route toward mega-pixel imaging by proving this architecture is viable with fewer phase shifters overall than conventional designs.

Kai: It’s about making the entire setup more compact and less power hungry as we aim for those larger scales in the future.

Lev: I’m looking forward to seeing how they tackle that hardware overhead because that's where most of the real-world complexity lands when you try to build this on a chip.

Mira: So, while it’s not quite at full mega-pixel imaging yet because of the single wavelength constraint, this paper shows the fundamental scaling behavior we need to pursue.

Kai: It’s a solid demonstration of how non-redundant antenna arrangements can overcome those traditional resolution bottlenecks in OPA systems.

Mira: This work gives us a clearer path for designing next-generation AI vision systems that need that fine spatial detail without needing the time and complexity of phase calibration for every frame.

Lev: I’m curious to see their next steps because if they can actually make this quadratic scaling robust with thousands of antennas, it changes what we can achieve in high-density information technology.

Kai: Exactly, so they’ve proven the concept works at this level, showing that we can get ten thousand resolvable points right now using this silicon photonic non-redundant optical phased array.

Mira: We'll keep an eye on their next work as they try to push past that single wavelength limitation and reduce the active components.

Lev: Next time, we’ll look at something completely different—maybe how we can use quantum entropy to measure these very complex systems they are talking about here.

School of Engineering, The University of Tokyo · Institute of Electromagnetic Fields, ETH Zurich · Sumitomo Electric Industries, Ltd. · Toyota Technological Institute

physics.optics

Submitted: 2026-08-21

Updated: 2026-08-21

Comments: 6 pages, 5 figures

Journal ref: Opt. Lett. 51 (2026) 6291-6294

DOI: 10.1364/OL.615011

License: http://creativecommons.org/licenses/by-nc-sa/4.0/

Importance score: 90/100

The gist: Optical phased arrays (OPAs) are promising wavefront controlling devices for imaging applications due to their compact and high-speed nature Here…

Key concepts

Non-Redundant Optical Phased Array (NR-OPA)
This is an antenna arrangement where every pair of antennas has a unique distance between them. Unlike conventional arrays with uniform spacing, this non-redundant layout allows the system to achieve quadratic scaling of spatial resolution with the number of antennas, maximizing information gain from the emitters.
Speckle-based Single-Pixel Imaging (SSPI)
This technique uses random electrical signals to drive N phase shifters in an OPA chip. The resulting interference creates speckle patterns that illuminate a target. A single photodetector measures the total light, and by switching illumination patterns K times, the target image is reconstructed using matrix inversion.
Resolution Scaling
For conventional OPAs with uniform spacing, resolution scales linearly with the number of antennas (N). In this NR-OPA system, due to its non-uniform antenna placement based on distinct displacement vectors, the theoretical maximum resolution scales quadratically with N (N^2 - N + 1), significantly boosting spatial fidelity.
Costas Array
The Costas array is the specific arrangement used for placing the grating couplers in this experimental chip. This arrangement dictates a precise grid pitch (15 µm) and determines the field of view achieved at a given wavelength, ensuring controlled light emission from each antenna element.

Terminology

Summary

Optical phased arrays (OPAs) are promising wavefront controlling devices for imaging applications due to their compact and high-speed nature Here, they experimentally demonstrate high-resolution imaging using a non-redundant OPA (NR-OPA) Due to the non-redundant antenna layout of NR-OPA based on the Costas array, the resolution scales quadratically with N Combined with the speckle-based single-pixel imaging (SSPI) scheme, which avoids the need for precise phase calibration, they experimentally achieve a large number of resolvable imaging points exceeding 10,000 using a silicon photonic NR-OPA chip with only N = 127 without sweeping the wavelength The demonstrated scheme provides a promising route toward mega-pixel imaging by a compact OPA chip with reduced number of phase shifters. p. 0

Where the paper says it

Due to the non-redundant antenna layout of NR-OPA based on the Costas array, the resolution scales quadratically with N. Combined with the speckle-based single-pixel imaging (SSPI) scheme, which avoids the need for precise phase calibration, we experimentally achieve a large number of resolvable imaging points exceeding 10,000 using a silicon photonic NR-OPA chip with only N = 127 without sweeping the wavelength.

Concept of Speckle-based Single-Pixel Imaging

The SSPI system involves splitting the input light into N waveguides within the OPA chip Light passing through each waveguide is phase controlled by a phase shifter and emitted from a grating antenna By driving N phase shifters of the OPA using random electrical signals, speckle-like far field patterns (FFPs) are generated by interference of the light from N emitters on the OPA. The generated intensity patterns are used to illuminate the imaging target, and the total optical power transmitted from the target is detected by a single-pixel photodetector (PD). p. 1

Where the paper says it

By driving N phase shifters of the OPA using random electrical signals, speckle-like far field patterns (FFPs) are generated by interference of the light from N emitters on the OPA. The generated intensity patterns are used to illuminate the imaging target, and the total optical power transmitted from the target is detected by a single-pixel photodetector (PD).

To retrieve the image, illumination patterns are switched for K times, and K PD signals are recorded. p. 1
Where the paper says it

The generated intensity patterns are used to illuminate the imaging target, and the total optical power transmitted from the target is detected by a single-pixel photodetector (PD). To retrieve the image, the illumination patterns are switched for K times, and K PD signals are recorded.

The k-th PD signal sk can be expressed using the transmittance distribution of the target object o and the k-th illumination pattern pk as sk = pk · o. p. 1
Where the paper says it

To retrieve the image, the illumination patterns are switched for K times, and K PD signals are recorded. The k-th PD signal sk can be expressed using the transmittance distribution of the target object o and the k-th illumination pattern pk as sk = pk · o.

Resolution Scaling and Theoretical Basis

For conventional periodic OPAs with antennas located on a uniform grid, the number of resolvable points scales only linearly with the number of antennas N To increase spatial resolution without sweeping the wavelength, it is effective to place optical antennas aperiodically with non-uniform spacing In NR-OPA, the antennas are located based on the concept of NRA where displacements between every pair of elements (r'ij in Fig. 1(c)) are mutually distinct. p. 1

Where the paper says it

In the NR-OPA, the antennas are located based on the concept of NRA, where the displacements between every pair of elements (r ′ ij in Fig. 1(c)) are mutually distinct.

Theoretically, the number of resolvable points can be evaluated by rank(P) which indicates the amount of independent information obtained about the object o from Eq. p. 1
Where the paper says it

An example of Costas array (N = 8) is displayed in Fig. Theoretically, the number of resolvable points can be evaluated by rank(P) [39], which indicates the amount of independent information obtained about the object o from Eq.

(1) and is equal to the number of distinct displacement vectors between emitters. p. 1
Where the paper says it

Theoretically, the number of resolvable points can be evaluated by rank(P) [39], which indicates the amount of independent information obtained about the object o from Eq. (1), and is equal to the number of distinct displacement vectors between emitters.

For NR-OPA, since these displacement vectors are mutually distinct, the imaging resolution is maximized to N(N − 1) + 1 = N squared − N + 1. p. 1
Where the paper says it

(1), and is equal to the number of distinct displacement vectors between emitters. For the case of NR-OPA, since these displacement vectors are mutually distinct, the imaging resolution is maximized to N(N − 1) + 1 = N2 − N + 1.

Experimental Implementation Details

The NR-OPA chip employed in this work consists of an edge coupler for a fiber input, a cascade of 1×2 multi-mode interference couplers (MMIs) to split the light to 128 waveguides (with one port used for monitoring), 127 thermo-optic phase shifters, and 127 grating couplers placed on a Costas array. The grid pitch of the Costas array was 15 µm, resulting in a field of view (FOV) of 5. p. 2

Where the paper says it

2(d)), and 127 grating couplers placed on a Costas array (Fig. The grid pitch of the Costas array was 15 µm, resulting in a field of view (FOV) of 5.

92◦ at a wavelength of 1550 nm. p. 2
Where the paper says it

The grid pitch of the Costas array was 15 µm, resulting in a field of view (FOV) of 5. 92◦ at a wavelength of 1550 nm.

Each grating coupler was fan-shaped, with a length of 10 µm and a width of 16 µm. p. 2
Where the paper says it

92◦ at a wavelength of 1550 nm. Each grating coupler was fan-shaped, with a length of 10 µm and a width of 16 µm.

The phase shifters were driven by pseudo-random voltage signals in a range from 2 V to 5 V such that the phases, which are proportional to the square of the voltages, were uniformly distributed. p. 2
Where the paper says it

The transmitted light from the imaging target was collected and the total power s was measured by an optical power meter (HP 81524A). The 127 phase shifters on the OPA were driven by pseudo-random voltage signals in a range from 2 V to 5 V, such that the phases, which are proportional to the square of the voltages, were uniformly distributed.

Performance Metrics and Results

The autocorrelation of the illumination pattern has a sharp peak at the center with full width at half maximum of 0.07 mm (equivalent to 0. p. 2

Where the paper says it

The autocorrelation has a sharp peak at the center with full width at half maximum of 0. 07 mm (equivalent to 0.

04◦), representing the fine-grained nature of the speckle. p. 2
Where the paper says it

07 mm (equivalent to 0. 04◦ ), which represents the fine-grained nature of the speckle.

The SVs for the experimentally obtained illumination matrix P remain relatively large up to ξ = 16, 003 (= N squared − N + 1), in good agreement with the simulated result. p. 2
Where the paper says it

They are then normalized by the largest SV. Unlike the uniform OPA, whose SV drop rapidly at ξ ∼ 500, the experimentally obtained SVs for the NR-OPA remain relatively large up to ξ = 16, 003 (= N2 − N + 1), in good agreement with the simulated result.

A reconstructed image of a target was obtained with a peak signal-to-noise ratio (PSNR) of 13. p. 2
Where the paper says it

The cutoff index of the truncated SVD was set to ξmax = 18, 000. The peak signal-to-noise ratio (PSNR) is 13.

9 dB. Cross-sectional plots confirm that line and space patterns with widths as narrow as 88. p. 3
Where the paper says it

5(b) and Fig. 5(d) that line and space patterns with widths as narrow as 88.

4 µm are successfully resolved. p. 3
Where the paper says it

5(d) that line and space patterns with widths as narrow as 88. 4 µm are successfully resolved, whereas the image starts to blur for narrower patterns as shown in Fig.

The total number of resolvable points is as large as (10 7 × 10 6) = 1. p. 4
Where the paper says it

imaging area is 9. 32 mm, the total number of resolvable points is as large as (107 × 106 =) 1.

13 × 10 4.

Conclusion

We have experimentally demonstrated high-resolution speckle-based OPA imaging at a single wavelength This is owing to the non-redundant antenna arrangement of our NR-OPA based on the Costas array, which enables the spatial resolution to scale quadratically with the number of antennas N. Such feature has distinct advantage over the conventional OPA with uniform spacings, where the resolution scales only linearly with N. p. 4

Where the paper says it

This is owing to the non-redundant antenna arrangement of our NR-OPA based on the Costas array, which enables the spatial resolution to scale quadratically with the number of antennas N. Such feature has distinct advantage over the conventional OPA with uniform spacings, where the resolution scales only linearly with N.

p. 4
Where the paper says it

This is owing to the non-redundant antenna arrangement of our NR-OPA based on the Costas array, which enables the spatial resolution to scale quadratically with the number of antennas N. Such feature has distinct advantage over the conventional OPA with uniform spacings, where the resolution scales only linearly with N.

By using a larger-scale NR-OPA with N ∼ 10 cubed, which is feasible with the state-of-the-art OPA technologies [3, 4], megapixel-order imaging should be realized. p. 4
Where the paper says it

Such feature has distinct advantage over the conventional OPA with uniform spacings, where the resolution scales only linearly with N. By using a larger-scale NR-OPA with N ∼ 103 , which is feasible with the state-of-the-art OPA technologies [3, 4], megapixel-order imaging should be realized.

Such feature has distinct advantage over the conventional OPA with uniform spacings, where the resolution scales only linearly with N.

REFERENCES

  1. C.-P. Hsu, B. Li, B. Solano-Rivas, et al. p. 5
    Where the paper says it

    Preprint 6 REFERENCES 1. Solano-Rivas, et al.

    , IEEE J. Sel. Top. Quantum Electron. p. 5
    Where the paper says it

    Solano-Rivas, et al. Quantum Electron.

    27, 8300416 (2021). p. 5
    Where the paper says it

    Quantum Electron. 27, 8300416 (2021).

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Non-redundant OPA hardware enables quadratic scaling of spatial resolution with antenna number N, allowing for mega-pixel imaging by a compact OPA chip with reduced number of phase shifters. This means AI systems can achieve significantly higher spatial resolution than conventional OPAs without wavelength sweeping.

  2. Speckle based single pixel imaging avoids the need for precise phase calibration, which is a major challenge in OPA-based systems. This allows AI image processing pipelines to operate without time-consuming phase calibration procedures, simplifying system design and deployment.

  3. The method reconstructs the target image using an equation derived from the signal and illumination patterns: o = P+s, where P+ is the Moore–Penrose pseudoinverse of P. This provides a direct mathematical framework for AI algorithms to perform image reconstruction from raw sensor data.

  4. The system can retrieve images with 10,000 resolvable points using only N = 127 antennas without wavelength sweeping. This capability enables high-resolution feature extraction in AI vision tasks that require fine spatial detail, such as detailed object recognition or defect detection at the micro-scale.

  5. The non-redundant antenna layout based on the Costas array ensures that the imaging resolution is maximized to N(N − 1) + 1 = N2 − N + 1. This quadratic scaling allows AI models to analyze images with vastly superior spatial fidelity compared to linearly scaling uniform OPA systems.

Abstract

Optical phased arrays (OPAs) are promising wavefront controlling devices for imaging applications due to their compact and high-speed nature. However, conventional periodic OPAs have a limited spatial resolution, which scales only linearly with the number of optical antennas N. Here, we experimentally demonstrate high-resolution imaging using a non-redundant OPA (NR-OPA). Due to the non-redundant antenna layout of NR-OPA based on the Costas array, the resolution scales quadratically with N. Combined with the speckle-based single-pixel imaging (SSPI) scheme, which avoids the need for precise phase calibration, we experimentally achieve a large number of resolvable imaging points exceeding 10,000 using a silicon photonic NR-OPA chip with only N=127 without sweeping the wavelength. The demonstrated scheme provides a promising route toward mega-pixel imaging by a compact OPA chip with reduced number of phase shifters.

Related papers