DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1-8, 10, 17 and 20-23 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hoydis et al1 (“Hoydis”) in view of Learning Orientation Notebook2 (“Learning Orientation”) and Jakob et al3 (“Jakob”).
Regarding claim 1, Hoydis teaches a method for computing a radio map (note the preamble does not add anything the body of the claim below does not address, such that addressing the limitations of the body of the claim addresses any limitations of the preamble; thus see Hoydis as explained below), comprising:
initializing parameters associated with a three-dimensional (3D) radio wave propagation environment (here any setting or starting of values for at least two parameters, or setting of a single parameter for many data points corresponds to such initializing where parameters are any quantities or data characterizing a 3D radio wave propagation environment; see Hoydis, Abstract, teaching “a GPU-accelerated open-source library for link-level simulations based on TensorFlow. Its latest release (v0.14) integrates a differentiable ray tracer (RT) for the simulation of radio wave propagation” such that here the system discloses simulation of radio wave propagation and such simulation is done with respect to an environment in which radio wave propagation is simulated or analyzed where such environment is a 3D radio wave propagation environment as in section 2 and 2A, teaching “Sionna RT is a ray tracing extension for radio propagation modeling” where “Sionna RT relies on Mitsuba 3 for the rendering and scene handling, e.g., its XML-file format, as well as the computation of ray intersections with scene primitives, i.e., triangles forming a mesh modeling a surface. The transformations of the polarized field components at each point of interaction between a ray and a scene object, e.g., reflections, are computed in TensorFlow, which is also used to combine the retained paths into (optionally) time varying CIRs” and “CIRs and functions thereof are differentiable with respect to many ray tracing parameters, including material properties, antenna patterns, orientations, and positions” and “Scene files for Mitsuba 3 may be created, edited, and exported using the popular open-source 3D content creation suite Blender [26] and the Mitsuba-Blender add-on [27]. The Blender-OSM add-on [28] allows one to rapidly create realistic scenes for almost any place in the world from OpenStreetMap [25]. In Sionna, scenes and radio propagation paths can be either rendered through the lens of configurable cameras via ray tracing or displayed with an integrated 3D viewer. For more detail on scene creation and rendering, we refer to Sionna’s documentation” and “Sionna RT allows for the definition of arbitrary radio materials which are characterized by their relative permittivity εr and conductivity σ” and “all transmitters and all receivers in a scene have the same antenna array configurations. For example, all transmitters may be equipped with an 8×2 dual-polarized array, while all receivers have a single cross-polarized antenna. Antenna arrays can be either explicitly modeled, i.e., paths are traced for every antenna element, or modeled synthetically after the ray tracing process by making a plane-wave assumption across the array” such that here parameters correspond to “material properties, antenna patterns, orientations, and positions” which are associated with a 3D radio wave propagation environment as these are set to model a 3D radio wave propagation environment such as any “scenes” loaded; see further section IIIB teaching “”);
tracing, by a ray tracer, paths representing an electric field originating at a transmitter (see Hoydis, section 1 teaching “differentiable ray tracer for radio propagation modeling” and as in section II, “Sionna RT is a ray tracing extension for radio propagation modeling” and “computation of ray intersections with scene primitives, i.e., triangles forming a mesh modeling a surface” and “transformations of the polarized field components at each point of interaction between a ray and a scene object, e.g., reflections are computed” and “CIRs and functions thereof are differentiable with respect to many ray tracing parameters” where the tracing of the ray tracer component is described in section IIB teaching “Reflected paths are computing using the image method [30] by either exhaustive search…or by a stochastic method” where the “stochastic method shoots a selected number of rays in random directions from the transmitter which then propagate through the scene until the maximum number of reflections has been reached” and as in section IIIB and figure 5, “differentiable ray tracing can be used to optimize the orientation of a transmit array in order to maximize the average received signal power in a specific region of a scene” and they “place a transmitter on a building within the example scene…and compute a coverage map for a small region…indicated by an orange ring” where “gradients of the average power received in this region with respect to the orientation of the transmitter are computed and used to optimized the latter via gradient ascent” such that here the transmitter is understood to be the source of the power received at any location which corresponds to the power of the radio signal modeled as being shot from the transmitter location into the scene where the ray is traced from the transmitter location/orientation to scattering surfaces of the 3D scene based on the set number of bounces allowed to be modeled) through a measurement surface, wherein a radio map associated with the measurement surface is computed based on interactions with scattering surfaces in the 3D radio wave propagation environment that are intersected by the paths (note that “interactions with scattering surfaces” does not specifically define the interaction and does not specifically define scattering such that if a surface is capable of altering a path then such a surface can be considered a scattering surface with altered ray corresponding to a scattered ray; see Hoydis as explained above teaching the tracing of paths originating at a transmitter, where such paths are traced through a measurement surface such as can be seen in figure 5 where a coverage map is depicted as a surface situated within the 3D scene, on which values of path gain are collected and displayed by location on a dB scale, where close inspection shows that that surface is a plane lying at a fixed elevation where buildings extend upward through it, where as further explained in section IIIB and as seen in figure 5, “the average received power within a small region of the scene” is computed such that this corresponds to the power of the path representing the electric field coming from the transmitter meaning that such a region and the plane inserted into the 3D scene correspond to a measurement surface as they measure the contribution of the traced paths through the measurement surface at that point; note that as explained above and with reference to section IIB, the “method shoots a selected number of rays in random directions from the transmitter which then propagate through the scene until a maximum number of reflections has been reached” and as in section IIA, there is a “maximum number of interactions between a ray and scene objects” which is set when tracing rays from the transmitter such that this corresponds to the radio map which is the visualized version of the coverage map and is computed based on such interactions with scattering surfaces in the 3D environment that are intersected by the paths);
evaluating a loss function associated with the radio map (see Hoydis, section IIIB, teaching “differentiable ray tracing can be used to optimize the orientation of a transmit array in order to maximize the average received signal power in a specific region of the scene” and to “compute a coverage map for a small region located behind the Arc de Triumphe” and “gradients of the average received power in this region with respect to the orientation of the transmitter are computed and used to optimize the latter via gradient ascent” such that here the maximizing of this objective is equivalent to evaluating a loss function associated with radio map in the form of the portion of the coverage map);
computing gradients of the loss function corresponding to the computed radio map (see Hoydis, section IIIB as explained above teaching as in section IIIB, “differentiable ray tracing can be used to optimize the orientation of a transmit array in order to maximize the average received signal power in a specific region of the scene” and to “compute a coverage map for a small region located behind the Arc de Triumphe” and “gradients of the average received power in this region with respect to the orientation of the transmitter are computed and used to optimize the latter via gradient ascent” such that here gradients of the average received power in that region, which is the objective being optimized, are computed such that these are gradients of the loss function which correspond to the computed radio map giving the objective being optimized).
Hoydis teaches all of the above but fails to explicitly teach replaying the paths to accumulate the gradients with additional gradients computed at the scattering surfaces, producing accumulated gradients corresponding to at least one of the parameters. As explained above, Hoydis teaches producing gradients in the computing step where these correspond to the radio map as explained above, but does not explicitly describe the additional gradients computed at the scattering surfaces and accumulating the gradients of the loss function with additional gradients computed at the scattering surface. Hoydis does teach that gradient computations are done with “TensorFlow’s automatic gradient computation” (see Hoydis, section II) but does not explicitly provide what such computations are. As will be shown below, Hoydis’ computation of the gradients of the loss function as explained above are actually accumulated with additional gradients computed at the scattering surfaces to produce accumulated gradients corresponding to at least one of the parameters, however, Hoydis does not explicitly disclose this in a single prior art reference, rather providing such details in the following reference as explained. Thus Hoydis fails to explicitly teach to accumulate the gradients with additional gradients computed at the scattering surfaces, producing accumulated gradients corresponding to at least one of the parameters. Additionally Hoydis does not teach replaying of the paths to accumulate the gradients.
In the same field of endeavor relating to ray tracing for radio propagation modeling, Learning Orientation is a notebook explaining the section IIIB example from Hoydis on “Optimization of Transmitter Orientation”, and teaches more specifically that that the optimization or learning of the parameter with respect to the coverage map generated comprises accumulating the gradients of the loss function with additional gradients computed at the scattering surfaces, producing accumulated gradients corresponding to at least one of the parameters (see Learning Orientation, pages 2-4, teaching “compute and show a coverage map for the transmitter “tx”” and “coverage map corresponds to the sum of squared amplitudes of the channel coefficients that fall into small rectangular regions (or cells) of a plane, which by default, is parallel to the plane, and sized such that it covers the entire scene with an elevation of 1.5m” where this corresponds to the radio map as explained above where the shooting and bouncing of rays as in Hoydis provides the values for the coverage map with respect to a transmitter emitting rays into the environment which are then traced according to the number of bounces set and for example page 3 teaches “white areas were not hit by any ray” meaning that rays are shot and bounced to get such information and page 6 explains “specular reflections are considered in this notebook” and as in “In [9]” on page 4, to “optimize the coverage of the target area through gradient descent with respect to the orientation of the transmitter” then it can be seen that TensorFlow is used and comprises use of “tf.GradientTape()” which a person having ordinary skill in the art recognizes as recording all computations that take place with regard to the taping which includes the computations in “scene.coverage_map” which involves recording the interactions at scattering surfaces such as the type of operation and the value recorded in that forward pass, and this is followed after the tape by “grads=tape.gradient(loss, tape.watched_variables())” which is understood by one having ordinary skill in the art as starting at the loss calculated and going backward along the tape to accumulate gradients at each recorded interaction such that here because the coverage map recorded on the tape of Learning Orientation is generated by shooting and bouncing rays from the transmitter and bouncing them off surfaces of the scene, and because the transformations applied to the field at each such bounce is performed within that same differentiable computation, the operations recorded on that tape include the transformation applied at each interaction of a path with a scattering surface, such that traversing that tape in reverse using tape.gradient produces a derivative value at each recorded interaction and those values are added together with the derivatives obtained at the values of the coverage map to ultimately provide the single gradient such that those per-interaction derivate values correspond to additional gradients at scattering surfaces and their combination into the single returned quantity with the map level gradients is the accumulation of the gradients into the returned quantity related to the orientation such that that is the accumulated gradients corresponding to at least one of the parameters).
Therefore it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to combine Hoydis and Learning Orientation to arrive at the claimed invention (other than the path replay limitation addressed below) as doing so would be no more than combining prior art elements according to known methods to yield predictable results. This is because first, the prior art contains the elements claimed as explained above, although not necessarily in a single prior art reference, with the only difference between the claimed invention and the prior art being the lack of actual combination of the elements in a single prior art reference. Furthermore, one of ordinary skill in the art could have combined the elements as claimed by known methods, and that in combination each element merely performs the same function as it does separately. Here it is clear that the Learning Orientation notebook describes the operation in Hoydis such that not only would one having ordinary skill in the art have been able to combine the teachings, it is already shown to have been combined and the functions performed in combination are the same as performed separately. Finally the results would have been predictable to one having ordinary skill in the art given that as explained above the Learning Orientation teachings describe the operation already explained as being carried out in Hoydis such that the results are predictable.
Hoydis as modified teaches all that is required as explained above, and teaches the paths to accumulate the gradients with additional gradients…producing accumulated gradients corresponding to at least one of the parameters as these paths correspond to the interactions recorded with regard to the coverage map used in the optimization of the transmitter orientation parameters as explained above. However, Hoydis as modified fails to teach or suggest replaying the paths to accumulate the gradients, where replaying a path corresponds to another run through the paths in some manner for the purposes recited. Instead Hoydis teaches the automatic differentiation above which utilizes the gradient tape function to store information about the interactions and utilizes the taped information to walk backwards accumulated gradients from interactions and based on the loss function and only traverses the paths in the forward pass a single time. Thus Hoydis as modified stands as a base device upon which the claimed invention can be seen as an improvement through such replaying of the paths to accumulate the gradients.
In the same field of endeavor relating to differentiable rendering and ray tracing of environments, Jakob teaches that making and keeping records such as the gradient tape is a problem in differentiating a ray-traced transport simulation (see Jakob, section 4.1 teaching “loops” corresponding to path tracing and encountering scattering media “are unfortunately a nuisance during differentiation, as each iteration generates intermediate state that must be reconstituted to compute reverse-mode derivatives (section 2.2)” and for a “path tracer, millions of Monte Carlo samples will run the loop in parallel for an unpredictable number of interactions” such that “large quantities of memory would need to be provisioned conservatively to store the resulting intermediate state” and “the automatic derivative of a loop is almost never satisfactory; the user should instead contribute domain-specific knowledge to implement an equivalent and more efficient custom adjoint”). As noted above, Hoydis as modified is a system which also shoots rays from a transmitter and bounces them through a scene and obtains its gradients by traversing a record of that computation and thus is the type of base system that could be improved by a custom adjoint.
Jakob further teaches a known technique for obtaining the same derivatives without traversing a record such as the tape of Hoydis as modified, and teaches replaying paths to accumulate gradients with additional gradients computed at scattering surfaces , producing accumulated gradients corresponding to parameters of the simulation (see Jakob, section 4.1, teaching “Path Replay Backpropagation (PRB)” which uses a “two-pass approach” where a “first pass generates a set of Monte Carlo samples and stores data consumed by a subsequent pass” and the “second pass regenerates the same set of samples, exploiting the precomputed data and arithmetic invertibility to recover incident radiance at ever vertex” and “With this information, the desired gradients can be accumulated”, such that here the sample is a traced path which recovers incident radiance at every vertex encountered so the second pass is traversing the same set of paths the first pass traced which is a replaying of the paths; with regard to the accumulation of the gradients then Jakob further teaches that the derivative values at the output are formed first and carried into the second pass as Jakob teaches in section 4.1, “differentiating the image loss initially yields adjoint radiance in image space, where it describes how pixels in the rendered image should change to reduce the loss” and that radiance in image space is the collection of derivative values of the loss taken with respect to the values accumulated at the output surface and correspond to the gradients of the loss function required by the claim and accumulation occurs and during the second pass a derivative value is added into the running total for a parameter at each surface interaction reached as for example “whenever adjoint radiance encounters a surface with differentiable parameters, the method accumulates a contribution into the local parameter gradient”; Jakob further teaches in figure 8 and its descriptions show a path-tracing kernel that scatters weighted sample values into an output buffer that is “passed into a loss function” and another kernel “contains adjoint versions of all steps and is responsible for accumulating scene parameter gradients” such that this path replay propagation technique corresponds to path replay to accumulate gradients as recited). Thus Jakob teaches a known technique applicable to the base system of Hoydis as modified.
Therefore it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to modify Hoydis as modified by applying the known technique of Jakob to implement the path replay backpropagation in the Hoydis as modified system as doing so would be no more than application of a known technique to a base device ready for improvement where the modification would yield predictable results and result in an improved system.
The predictable result of such a combination would be that the gradients in Hoydis as modified would be obtained by replaying the paths traced as taught by Jakob, instead of traversing a record made during the forward computation as in Hoydis as modified. The forward computation in Hoydis as modified would remain the same, but gradients would be accumulated based on the path replay backpropagation (PRB) of Jakob. The results would be predictable because the computations in Jakob’s PRB have the same structure as in Hoydis as modified where each accumulate weighted contributions of paths traced from a source onto an output surface, and reduce values related to that surface to a single objective and then work backward from that objective to some parameter, so that the sequence of dependent computations traversed by the replay is the same in both. The combination would also result in an improved system as it would obtain the same gradients as in Hoydis but would not require that the intermediate values produced at every interaction of every traced ray be retained in memory in order to obtain them as taught and suggested by Jakob (see Jakob, section 2.2 teaching exhaustive storage, such as utilized by Hoydis as modified when using traditional automatic differentiation such as provided in TensorFlow and as seen in figure 10 for example, is not ideal where “exhaustive storage is simple but does not scale, as modern processors can generate many terabytes of intermediate state per second” and thus “adjoints” for such calculation should be used such as PRB where as in section 4.1 it is explained that using a reverse mode automatic differentiation in situations involve path tracing and Monte Carlo Sampling “will run the loop in parallel for an unpredictable number of interactions” such that “large quantities of memory would need to be provisioned conservatively to store the resulting intermediate state” and “the automatic derivative of a loop is almost never satisfactory; the user should instead contribute domain-specific knowledge to implement an equivalent and more efficient custom adjoint”).
Regarding claim 2, Hoydis as modified teaches all that is required as applied to claim 1 above and further teaches wherein during the tracing intermediate data associated with the interactions is not stored to a memory (see Hoydis as modified where the PRB technique already combined from Jakob teaches such limitations as the advantages of replaying the paths where intermediate data associated with the interactions is not stored to a memory but rather the first pass does not store the per-interaction state but instead stores a limited quantity of per-path data where as in section 4.1, the pass “generates a set of Monte Carlo samples and stores data consumed by a subsequent pass” such as in figure 8 teaching a “per-sample state array” and such intermediate data is not retained during the tracing as is also shown by the second pass needing to recover such intermediate data in the replay) and the gradients of the loss function are stored to the memory before replaying the paths (see Hoydis as modified where the PRB technique already combined from Jakob teaches such limitations as the advantages of replaying the paths where intermediate data associated with the interactions is not stored to a memory but rather the first pass does not store the per-interaction state but instead stores a limited quantity of per-path data where as in section 4.1, the pass “generates a set of Monte Carlo samples and stores data consumed by a subsequent pass” such as in figure 8 teaching a “per-sample state array” and such intermediate data is not retained during the tracing as is also shown by the second pass needing to recover such intermediate data in the replay and the gradients of the loss function are stored to memory before replaying the paths as section 4.1 teachings that “differentiating the image loss initially yields adjoint radiance in image space, where it describes how pixels in the rendered image should change to reduce the loss” such as in figure 8 where those values are carried from the kernel in which the loss is evaluated to the adjoint kernel in which the parameter gradients are accumulated, such that when it is formed in one pass and consumed during a later pass this means those values are in memory in the time between) and the intermediate data is recomputed during the replaying to compute the gradients (see Hoydis as modified where the PRB technique already combined from Jakob teaches such limitations as in section 4.1 teaching “this second pass regenerates the same set of samples, exploiting the precomputed data and arithmetic invertibility to recover the incident radiance at every vertex” and “With this information, the desired gradients can be accumulated” such that here recovering a quantity at every vertex of a replayed path is the recomputing during the replaying of the intermediate data of the interactions and the accumulation of the gradients from the information that is recovered is the computation of the gradients from that).
Regarding claim 3, Hoydis as modified teaches all that is required as applied to claim 1 above and further teaches updating the parameters using the accumulated gradients (see Hoydis, section IIIB teaching updating the transmitter orientation parameters using the accumulated gradients where they are updated according to the optimization where they are “used to optimize the latter via gradient descent” and this can be seen in Learning Orientation on page 4 teaching “optimizer.apply_gradients” which outputs the resulting change in the value of that parameter over the course of the optimization such that for example the orientation starts at 0,0,0 and is updated during the optimization using those accumulated gradients as explained above).
Regarding claim 4, Hoydis as modified teaches all that is required as applied to claim 1 above and further teaches the electric field comprises polarization state and phase (see Hoydis as modified where Hoydis teaches that the electric field comprises polarization state as the quantities transformed at each interaction are the polarized components of the field as in Hoydis, section II teaching “transformation of the polarized field components at each point of interaction between a ray and a scene object, e.g., reflections, are computed in TensorFlow” and transmitters are modeled to emit electric fields modeled as “8x2 dual-polarized array” for example and the electric field comprises a phase given that phase shifts are applied to the traced paths in computing the channel impulse responses where “Phase shifts related to the relative antenna positions will then be applied based on a plane-wave assumption when the CIRs are computed”).
Regarding claim 5, Hoydis as modified teaches all that is required as applied to claim 1 above and further teaches the parameters comprise at least one of a meta material, an antenna pattern, an antenna orientation, an antenna position, scene geometry, configuration of reconfigurable intelligent surfaces and meta materials, a configurable reflective surface, array geometry, Doppler map, and transmitter and receiver directivity, orientations, and positions (see Hoydis, section IIIB teaching the parameter is transmitter orientation that is being optimized).
Regarding claim 6, Hoydis as modified teaches all that is required as applied to claim 1 above and further teaches wherein the radio map comprises at least one of a path loss map, root mean squared delay spread map, direction spread of arrival map, and direction spread of departure map (see Hoydis, figure 2 and section IIIB and figure 5 teaching a “coverage map” which is a path loss map as a map of path gain such as “Visualization of the coverage map (path gain [dB])” shows a scale running from 0 to -120 dB and given that path loss and path gain are the same quantity just expressed with different signs, then the path gain at any point also shows a path loss of the opposite sign at the same location making them functionally the same).
Regarding claim 7, Hoydis as modified teaches all that is required as applied to claim 1 above and further teaches wherein the parameters for which the gradients are computed configure at least one of material properties of the 3D radio wave propagation environment, geometry of the 3D radio wave propagation environment, and the transmitter (see Hoydis as modified where Hoydis as in section IIIB teaches the gradients are computed to configure the transmitter and optimize its orientation through learning via gradient descent).
Regarding claim 8, Hoydis as modified teaches all that is required as applied to claim 1 above and further teaches wherein the measurement surface comprises a grid of cells including a first cell and a second cell, the first cell having a first surface area that differs from a second surface area of the second cell (see Hoydis as modified where as in Learning Orientation as already combined it is explained that “coverage map is divided into cells of size cm_cell_size (in meters)” and “each cell is 2mx2m” such that each cell has a first surface area corresponding to the region it covers and thus while the total surface area of each area is the same, the first and second cells have first and second surface areas that differ in that each represents a different portion of the measurement such that their surface areas differ in that manner, as for example a ray may enter or pass through the surface area of a first cell without entering or passing through the surface area of the second cell).
Regarding claim 10, Hoydis as modified teaches all that is required as applied to claim 1 above and further teaches wherein the radio map comprises a grid of cells and each cell is mapped to a vector (see Hoydis as modified teaching as explained in Learning Orientation that the radio or coverage map is “divided into cells of size cm_size” and the coverage map as divided then is an array in which the cell coordinates are two of the indices and the transmitter is a further index such as where there is coverage map with values (1, 339, 427) which corresponds to the number transmitter and the number of the cell in the x and y direction such that each cell of the grid is mapped to an ordered, indexed set of values within that array where such set of values may be considered a vector mapped to each cell).
Regarding claim 17, Hoydis as modified teaches all that is required as applied to claim 1 above and further teaches wherein at least one of the steps of initializing, tracing, evaluating, computing, or replaying is performed for training, testing, or certifying a neural network employed in a machine, robot, or autonomous vehicle (see Hoydis as modified where Hoydis in section I teaches “research topics require the simulation of specific radio environments by ray tracing” such as “machine-learning (ML)-based transceiver algorithms” because “a spatially consistent correspondence between a physical location in a scene and the channel impulse response (CIR) is required which the widely used stochastic channel models…cannot provide” which is why Hoydis “added a differentiable ray tracing module (RT)” to its “simulator” and others use “neural networks (NNs) for path loss prediction” or “attempts to model ray-surface interactions by NN’s” and the technique is created for “fusion of ML and ray tracing” meaning that as the steps performed in Hoydis as modified are performed such as relating to determining path loss prediction in an environment being model then this means they are performed for training, testing, or certifying a neural network employed in a machine running the neural network (some machine of course being inherently necessarily) such as would be providing path loss predictions or the like or ray-surface interactions “by NNs”).
Regarding claims 20-21, the instant claims correspond to a “system, comprising: a memory that stores a radio map; and a processor that is connected to the memory, wherein the processor is configured to compute the radio map” and wherein all of the functions recited correspond to the functions of claims 1-2, respectively. Hoydis as modified teaches such a system where a memory is required to store the radio map computed such that it can be operated on and a GPU or CPU as in Hoydis provides the computation. In light of this, the limitations of claims 20-21 correspond the limitations of claims 1-2, respectively; thus they are rejected on the same grounds as claims 1-2, respectively.
Regarding claims 22-23, the instant claims correspond to an apparatus in the form of a “non-transitory computer-readable media storing instructions for computing a radio map that, when executed by one or more processors, cause the one or more processors to perform the steps” such as those recited in claims 1 and 4. Hoydis as modified as explained in the rejection of claim 1 above is a software program that is embodied on a computer-readable media as can be seen by the software code describing the functions which are carried out which would be by a processor as described in Hoydis as modified. In light of this the limitations of claims 22-23 correspond to the limitations of claims 1 and 4, respectively; thus they are rejected on the same grounds as claims 1 and 4, respectively.
Claim(s) 15-16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hoydis as modified as applied to claim 1 above, and further in view of Corgan et al4 (“Corgan”).
Regarding claim 15, Hoydis as modified teaches all that is required as applied to claim 1 above but fails to specifically teach wherein at least one of the steps of initializing, tracing, evaluating, computing, or replaying is performed on a server or in a data center and the computed radio map is streamed to a user device. Rather Hoydis teaches the computations such as initializing, tracing, evaluating, computing, or replaying are performed with respect to GPUs but does not limit any of the computations to a specific environment. Furthermore, while Hoydis also teaches computing a radio map and presenting it to a user such as can be seen in the Hoydis and Learning Orientation as explained above, it is not disclosed as limited to a particular situation of streaming or providing data from a server source to a user device. Thus Hoydis as modified stands as a base device upon which the claimed invention can be seen as an improvement through utilizing the basic server/client architecture to perform computations at a remote source which are then streamed or sent to a user device.
In the same field of endeavor relating to simulating radio wave propagation in an environment (see Corgan, paragraph 0048 teaching “training, technical details, and use of radio frequency radiance field (RF-RF) models for RF signal transmission and reception control, RF communication simulations, RF network deployment, and other purposes. These models encode RF signal propagation characteristics of an environment for efficient calculation of wireless channel characteristics in the environment. Using trained RF-RF models, various network devices, such as user equipment (UE), cellular phones, vehicles, base stations, and/or back-end network systems can optimize their RF control to provide improved and more-efficient communication” and “wireless propagation and channel modeling are critical tools for the development and optimization of wireless systems. By predicting the characteristics of propagation of wireless signals through an environment-that is, predicting characteristics of a wireless channel-transmission, reception, coverage modeling, shared use, and general deployment and optimization of wireless systems for communications, radar, and other applications can be optimized”), Corgan teaches that it is known to perform radio propagation steps similar to initializing, tracing, evaluating, computing, or replaying on a server or in a data center (see Corgan, paragraphs 0177-0179 teaching “a computing system that may be used to implement one or more components of a system that utilizes RF-RF models for RF system operations. The computer system illustrated in FIG. 15 can be, or can include, one or more of the network devices and modules described herein, e.g., UE, DU, RU, CU, and cloud computing system, for example in any of systems 100, 600, and/or 700. The computer system illustrated in FIG. 15 , and/or a component or portion thereof, can be used to perform any of the processes described herein, such as processes 200, 610, 800, 900, 1000, 1100, 1200, and/or 1300” and “computing system includes computing device 1500 and a mobile computing device 1550 that can be used to implement the techniques described herein. For example, either or both of the computing device 1500 and the mobile computing device 1550 can execute an RF-RF model for RF communication control and/or other purposes” and “computing device 1500 is intended to represent various forms of digital computers and network components, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, cloud computing systems, base stations, mainframes, back-end network equipment, and other appropriate computers. The mobile computing device 1550 is intended to represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smart-phones, mobile embedded radio systems, radio diagnostic computing devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to be limiting” such that here computations that an end user will use at a user device can be carried out on a server or in a data center) and the computed information is streamed to a user device (see Corgan, paragraphs 0177-0179 teaching the calculations carried out with respect to the server and as in paragraph 0195 the “Implementations can include a back end component, e.g., a data server, or a middleware component, e.g., an application server, or a front end component, e.g., a client computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the subject matter described is this specification, or any combination of one or more such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (“LAN”) and a wide area network (“WAN”), e.g., the Internet” such that here a front-end component provides the function to stream computed results from a server backend to a user device). Thus Corgan teaches a known technique applicable to the base system of Hoydis as modified.
Therefore it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to modify Hoydis as modified by applying the teachings of Corgan as doing so would be no more than application of a known technique to a base system ready for improvement which would yield predictable results and result in an improved system. The predictable result of the combination would be that the computations in Hoydis as modified are carried out using a server or data center and the results are sent to a front end to a user for display of the output just as suggested with the information in Corgan. This would result in an improved system as the system would be able to take advantage of a server or data center that unburdens a user device from having the same processing power.
Regarding claim 16, Hoydis as modified teaches all that is required as applied to claim 1 but fails to specifically teach wherein at least one of the steps of initializing, tracing, evaluating, computing, or replaying is performed within a cloud computing environment. Thus Hoydis as modified stands as a base device upon which the claimed invention can be seen as an improvement through utilizing some cloud computing environment to perform computations.
In the same field of endeavor relating to simulating radio wave propagation in an environment (see Corgan, paragraph 0048 teaching “training, technical details, and use of radio frequency radiance field (RF-RF) models for RF signal transmission and reception control, RF communication simulations, RF network deployment, and other purposes. These models encode RF signal propagation characteristics of an environment for efficient calculation of wireless channel characteristics in the environment. Using trained RF-RF models, various network devices, such as user equipment (UE), cellular phones, vehicles, base stations, and/or back-end network systems can optimize their RF control to provide improved and more-efficient communication” and “wireless propagation and channel modeling are critical tools for the development and optimization of wireless systems. By predicting the characteristics of propagation of wireless signals through an environment-that is, predicting characteristics of a wireless channel-transmission, reception, coverage modeling, shared use, and general deployment and optimization of wireless systems for communications, radar, and other applications can be optimized”), Corgan teaches that it is known to perform radio propagation modeling related computations within a cloud computing environment (see Corgan, paragraphs 0177-0179 teaching “a computing system that may be used to implement one or more components of a system that utilizes RF-RF models for RF system operations. The computer system illustrated in FIG. 15 can be, or can include, one or more of the network devices and modules described herein, e.g., UE, DU, RU, CU, and cloud computing system, for example in any of systems 100, 600, and/or 700. The computer system illustrated in FIG. 15 , and/or a component or portion thereof, can be used to perform any of the processes described herein, such as processes 200, 610, 800, 900, 1000, 1100, 1200, and/or 1300” and “computing system includes computing device 1500 and a mobile computing device 1550 that can be used to implement the techniques described herein. For example, either or both of the computing device 1500 and the mobile computing device 1550 can execute an RF-RF model for RF communication control and/or other purposes” and “computing device 1500 is intended to represent various forms of digital computers and network components, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, cloud computing systems, base stations, mainframes, back-end network equipment, and other appropriate computers. The mobile computing device 1550 is intended to represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smart-phones, mobile embedded radio systems, radio diagnostic computing devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to be limiting” such that here computations that an end user will use at a user device can be carried out on a server or in a data center). Thus Corgan teaches a known technique applicable to the base system of Hoydis as modified.
Therefore it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to modify Hoydis as modified by applying the teachings of Corgan as doing so would be no more than application of a known technique to a base system ready for improvement which would yield predictable results and result in an improved system. The predictable result of the combination would be that the computations in Hoydis as modified are carried out in a cloud computing environment just as suggested with the computations in Corgan. This would result in an improved system as the system would be able to take advantage of a cloud computing environment that unburdens a user device from having the same processing power.
Claim(s) 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hoydis as modified as applied to claim 1 above, and further in view of Subtil et al5 (“Subtil”).
Regarding claim 18, Hoydis as modified teaches all that is required as applied to claim 1 above but fails to specifically teach wherein at least one of the steps of initializing, tracing, evaluating, computing, or replaying is performed on a virtual machine comprising a portion of a graphics processing unit. Thus Hoydis stands as a base device upon which the claimed invention can be seen as an improvement.
In the same field of endeavor relating to ray and path tracing computations being performed on various hardware setups (see Subtil, paragraphs 0021-0024 teaching “ray or path tracing is an example of a technique used for rendering scenes in a rendering pipeline; especially when the scenes include complex lighting. Ray tracing describes any number of techniques used for efficiently resolving visibility along a straight line between any two arbitrary points in a scene, where the visibility information is used to resolve light transport and light interactions with materials in the scene” and “systems and methods of the present disclosure provide a renderer and a rendering process employing ray or path tracing and image-space filtering that interleaves the pixels of a frame into partial image fields and corresponding reduced-resolution images that are individually processed in parallel” and “parallel processing can be performed by a single processor or by multiple processors—such as different graphics processing unit (GPU) resources, including different threads of a single GPU, one or more discrete GPUs, one or more virtual GPUs (vGPUs), etc.—and is applicable to both local and remote computing”) and teaches computations relating to such ray and path tracing are performed on a virtual machine comprising a portion of a graphics processing unit (see Subtil, paragraphs 0021-0024 teaching as continued from above, “parallel processing can be performed by a single processor or by multiple processors—such as different graphics processing unit (GPU) resources, including different threads of a single GPU, one or more discrete GPUs, one or more virtual GPUs (vGPUs)” and for example “When employing two GPUs (e.g., discrete or virtual, such as part of a virtual machine), for example, each GPU can render alternate pixels (e.g., in each row or column) of the frame. The disclosed features can also be easily extended to systems with more GPUs available for parallel processing” such that here a virtual machine can be employed such as a vGPU and part of such a virtual machine may be used to perform computations). Thus Subtil teaches known techniques applicable to the base system of Hoydis as modified.
Therefore it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to modify Hoydis as modified such that at least one of the steps of initializing, tracing, evaluating, computing, or replaying is performed on a virtual machine comprising a portion of a graphics processing unit as doing so would be no more than application of a known technique to a base device ready for improvement which would yield predictable results and result in an improved system. The predictable result would be that the computations described in Hoydis as modified would be assigned to portions of a vGPU which would output the information instead of the GPU or CPU used in Hoydis as modified. This would result in an improved system as it would allow the user to utilize more powerful processing resources such as a vGPU and would free their own end user device from having to perform complex calculation such as those involved in ray tracing and give the traditional advantages of distributed computing as would be understood by one having ordinary skill in the art.
Allowable Subject Matter
Claims 9, 11-14 and 19 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
The following is a statement of reasons for the indication of allowable subject matter: the prior art of record fails to teach or suggest the respective claim limitations when considered as a whole.
Regarding claim 9, the instant claim requires wherein the measurement surface comprises either a volume partitioned into a grid of cuboids or a surface that is non-contiguous. Hoydis as modified fails to teach or suggest this limitation as rather Hoydis as modified partitions the measurement surface into a contiguous surface and does not teach or suggest partitioning it into a grid of cuboids. The Examiner is unable to find any teaching or suggestion of such a limitation in the relevant context of the claims. Thus the claim contains allowable subject matter.
Regarding claim 11, Hoydis as modified teaches all that is required as applied to claim 1 but fails to teach wherein the measurement surface is a non-planar surface in 3D. Rather Hoydis as modified only teaches that the measurement surface is a planar surface in 3D. The Examiner is unable to find any teaching or suggestion of such a limitation in the relevant context of the claims. Thus the claim contains allowable subject matter.
Regarding claim 12, Hoydis as modified teaches all that is required as applied to claim 1 above but fails to teach wherein at least one path of the paths intersects the measurement surface more than once and further comprising combining matrices corresponding to each intersection with the measurement surface to compute a transfer matrix corresponding to the interactions with the scattering surfaces. Rather while Hoydis as modified does teach paths intersect the measurement surface more than once as this could be a natural consequence of the shooting and bouncing of rays to compute the coverage map, but Hoydis as modified is silent as to any combining of matrices corresponding to each intersection with the measurement surface to compute a transfer matrix corresponding to the interactions with the scattering surfaces. The Examiner is unable to find any other teaching or suggestion of such a limitation in the prior art. Thus the claim contains allowable subject matter. Note that claims 13-14 are considered allowable at least because they are based upon a claim containing allowable subject matter.
Regarding claim 19, Hoydis as modified teaches all that is required as applied to claim 1 above but fails to teach or suggest wherein at least one of the steps of initializing, tracing, evaluating, computing, or replaying is implemented to include advanced error correction, fault-tolerance, and self-healing capabilities. Rather Hoydis as modified is silent as to any of the steps including each of advanced error correction, fault-tolerance, and self-healing capabilities. The Examiner is unable to find any other teaching or suggestion in the prior art of such limitations as applied to the invention described by the claims. Thus the claim contains allowable subject matter.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. See Vicini et al6 teaching the Path Replay Backpropagation technique described in the Jakob prior art in more detail above. Vicini teaches the path replay backpropagation as applied to illumination radiation but does not specifically teach radio wave propagation or adapting such types of modeled paths of radio waves in detail.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SCOTT E SONNERS whose telephone number is (571)270-7504. The examiner can normally be reached Mon-Friday 9-5.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Xiao Wu can be reached at (571) 272-7761. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/SCOTT E SONNERS/Examiner, Art Unit 2613
/XIAO M WU/Supervisory Patent Examiner, Art Unit 2613
1 Jakob Hoydis, Aoudia, F. A., Cammerer, S., Nimier-David, M., Binder, N., Marcus, G., & Keller, A. (20 Mar 2023). Sionna RT: Differentiable Ray Tracing for Radio Propagation Modeling (2303.11103v1). arXiv. https://arxiv.org/abs/2303.11103v1
2 NVlabs. (20 Mar 2023). diff-rt/Learning_Orientation.ipynb at 7c65d298ea93c3614b67aab46e361bb8770ced68 · NVlabs/diff-rt. GitHub. https://github.com/NVlabs/diff-rt/blob/7c65d298ea93c3614b67aab46e361bb8770ced68/Learning_Orientation.ipynb
3 Jakob W, Speierer S, Roussel N, Vicini D. Dr. jit: A just-in-time compiler for differentiable rendering. ACM Transactions on Graphics (TOG). 2022 Jul 22;41(4):1-9.
4 US PGPUB No. 2024/0323719
5 US PGPUB No. 20200372699
6 Vicini D, Speierer S, Jakob W. Path replay backpropagation: differentiating light paths using constant memory and linear time. ACM Transactions on Graphics (TOG). 2021 Jul 19;40(4):1-4.
Prosecution Insights