Fixed quantization alignment in machine learning based channel state information compression

Abstract

Systems, methods, apparatuses, and computer program products for fixed quantization alignment in machine learning based channel state information (CSI) compression. The method may include training, at a network device, at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme and determining quantization block information. The quantization block information may be based on the at least one fixed quantization setting. The method may also include sending, to a user equipment, encoder model information and obtaining channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

Description

TITLE:FIXED QUANTIZATION ALIGNMENT IN MACHINE LEARNING BASED CHANNELSTATE INFORMATION COMPRESSIONFIELD:

[0001] Some example embodiments may generally relate to mobile or wireless telecommunication systems, such as Long Term Evolution (LTE) or fifth generation (5G) radio access technology or new radio (NR) access technology, or other communications systems. For example, certain example embodiments may relate to apparatuses, systems, and / or methods for fixed quantization alignment in machine learning based channel state information (CSI) compression.BACKGROUND:

[0002] Examples of mobile or wireless telecommunication systems may include the Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access Network (UTRAN), LTE Evolved UTRAN (E-UTRAN), LTE- Advanced (LTE- A), MulteFire, LTE- A Pro, fifth generation (5G) radio access technology or NR access technology, and / or 5G- Advanced. 5G wireless systems refer to the next generation (NG) of radio systems and network architecture. 5G network technology is mostly based on NR technology, but the 5G (or NG) network can also build on E-UTRAN radio. It is estimated that NR may provide bitrates on the order of 10-20 Gbit / s or higher, and may support at least enhanced mobile broadband (eMBB) and ultra-reliable low-latency communication (URLLC) as well as massive machine-type communication (mMTC). NR is expected to deliver extreme broadband and ultra-robust, low-latency connectivity and massive networking to support the loT.

[0003] Mobile and wireless telecommunication systems often communicate channel state information. Channel state information is sometimes compressed before being communicated over the mobile or wireless network to reduce overhead of the communications.SUMMARY

[0004] An embodiment may be directed to an apparatus. The apparatus may include at least one processor and at least one memory storing instructions. The instructions stored in the at least memory, when executed by the at least one processor may cause the apparatus at least to perform training at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme and determining quantization block information. The quantization block information may be based on the at least one fixed quantization setting. The instructions may also cause sending, to a user equipment, encoder model information and obtaining channel state information by dequantizing a received compressed channel state information using the at least one fixed quantization setting.

[0005] Another embodiment may be directed to an apparatus. The apparatus may include at least one processor and at least one memory storing instructions. The instructions stored in the at least memory, when executed by the at least one processor may cause the apparatus at least to perform establishing a quantization block based, at least partially, on quantization block information. The instructions may also cause establishing an encoder based, at least partially, on received encoder model information from a network device and estimating channel state information of a channel. The instructions may also cause compressing the channel state information of the channel using the quantization block and the encoder and include sending, to the network device, the compressed channel state information.

[0006] An embodiment may be directed to a method. The method can include training, at a network device, at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme and determining quantization block information. The quantization block information may be based on the at least one fixed quantization setting. The method may also include sending, to a user equipment, encoder model information and obtaining channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

[0007] Another embodiment may be directed to a method. The method can include establishing, at a user equipment, a quantization block based, at least partially, on quantization block information. The method may include establishing, at the user equipment, an encoder based, at least partially, on received encoder model information from a network device and estimating channel state information of a channel. The method may also include compressing the channel state information of the channel using the quantization block and the encoder and include sending, to the network device, the compressed channel state information.

[0008] Another embodiment may be directed to an apparatus. The apparatus may include means for training at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme and means for determining quantization block information. The quantization block information may be based on the at least one fixed quantization setting. The apparatus may also include means for sending, to a user equipment, encoder model information and means for obtaining channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

[0009] Another embodiment may be directed to an apparatus. The apparatus may include means for establishing, at a user equipment, a quantization block based, at least partially, on quantization block information. The apparatus may include means for establishing an encoder based, at least partially, on received encoder model information from a networkdevice and estimating channel state information of a channel. The apparatus may also include means for compressing the channel state information of the channel using the quantization block and the encoder and means for sending, to the network device, the compressed channel state information.

[0010] Another embodiment may be directed to an apparatus comprising circuitry configured to perform a method. The method can include training at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme and determining quantization block information. The quantization block information may be based on the at least one fixed quantization setting. The method may also include sending, to a user equipment, encoder model information and obtaining channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

[0011] Another embodiment may be directed to an apparatus comprising circuitry configured to perform a method. The method can include establishing, at a user equipment, a quantization block based, at least partially, on quantization block information. The method may include establishing an encoder based, at least partially, on received encoder model information from a network device and estimating channel state information of a channel. The method may also include compressing the channel state information of the channel using the quantization block and the encoder and include sending, to the network device, the compressed channel state information.

[0012] Another embodiment may be directed to a non-transitory computer readable medium comprising program instructions stored thereon that, when executed by an apparatus, cause the apparatus to perform at least a method. The method can include training at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme and determining quantization block information. The quantization block information may be based on the at least one fixed quantization setting. The method may also include sending, to a user equipment, encoder model information andobtaining channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

[0013] Another embodiment may be directed to a non-transitory computer readable medium comprising program instructions stored thereon that, when executed by an apparatus, cause the apparatus to perform at least a method. The method can include establishing, at a user equipment, a quantization block based, at least partially, on quantization block information. The method may include establishing an encoder based, at least partially, on received encoder model information from a network device and estimating channel state information of a channel. The method may also include compressing the channel state information of the channel using the quantization block and the encoder and include sending, to the network device, the compressed channel state information.

[0014] A computer program comprising instructions, which, when executed by an apparatus, cause the apparatus to perform a method. The method can include training, at a network device, at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme and determining quantization block information. The quantization block information may be based on the at least one fixed quantization setting. The method may also include sending, to a user equipment, encoder model information and obtaining channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

[0015] A computer program comprising instructions, which, when executed by an apparatus, cause the apparatus to perform a method. The method can include establishing, at a user equipment, a quantization block based, at least partially, on quantization block information. The method may include establishing an encoder based, at least partially, on received encoder model information from a network device and estimating channel state information of a channel. The method may also include compressing the channel state information of the channel using the quantization block and the encoder and include sending, to the network device, the compressed channel state information.BRIEF DESCRIPTION OF THE DRAWINGS:

[0016] For proper understanding of example embodiments, reference should be made to the accompanying drawings, wherein:

[0017] FIG. 1 illustrates an example flow diagram of communications between a user equipment and a network element, according to certain example embodiments.

[0018] FIG. 2 illustrates another example flow diagram of communications between the user equipment and the network element, according to certain example embodiments.

[0019] FIG. 3 illustrates another example flow diagram of communications between the user equipment and the network element, according to certain example embodiments.

[0020] FIG. 4 illustrates an example flow chart for a method of controlling a network element, according to certain example embodiments.

[0021] FIG. 5 illustrates an example flow chart for a method of controlling a user equipment, according to certain example embodiments.

[0022] FIG. 6A illustrates an example vector quantization, according to certain example embodiments.

[0023] FIG. 6B illustrates another example vector quantization with 2D uniform distribution, according to certain example embodiments.

[0024] FIG. 7A illustrates an example vector quantization with 2D standard Gaussian distribution, according to certain example embodiments.

[0025] FIG. 7B illustrates another example vector quantization with 2D standard Gaussian distribution, according to certain example embodiments.

[0026] FIG. 8 illustrates a set of apparatuses, according to certain example embodiments.DETAILED DESCRIPTION:

[0027] It will be readily understood that the components of certain example embodiments, as generally described and illustrated in the figures herein, may be arranged and designed in a wide variety of different configurations. The following is a detailed description of some example embodiments of systems, methods, apparatuses, and computer program products for fixed quantization alignment in machine learning based channel state information (CSI) compression.

[0028] The features, structures, or characteristics of example embodiments described throughout this specification may be combined in any suitable manner in one or more example embodiments. For example, the usage of the phrases “certain embodiments,” “an example embodiment,” “some embodiments,” or other similar language, throughout this specification refers to the fact that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment. Thus, appearances of the phrases “in certain embodiments,” “an example embodiment,” “in some embodiments,” “in other embodiments,” or other similar language, throughout this specification do not necessarily refer to the same group of embodiments, and the described features, structures, or characteristics may be combined in any suitable manner in one or more example embodiments. Further, the terms “base station”, “cell”, “node”, “gNB”, “network” or other similar language throughout this specification may be used interchangeably.

[0029] As used herein, “at least one of the following: ” and “at least one of ” and similar wording, where the list of two or more elements are joined by “and” or “or,” mean at least any one of the elements, or at least any two or more of the elements, or at least all the elements.

[0030] FIG. 1 illustrates an example flow diagram 1000 of communications between a user equipment (UE) 100 and a network 200, according to certain example embodiments. The network (NW) 200 and the UE 100 may communicate electronically wirelessly, through wires, through fiber optic cables, or a combination of these and / or other methods of electronic communication. The wireless communication may be in a 5G network. The UE 100 may communicate directly or indirectly with the network 200. The UE 100 may communicate with a single network element in the network 200 or with multiple network elements. The network 200 may include base stations, servers, and other elements not shown in FIG. 1. The UE 100 may be a user device such as a cell phone, laptop, computer, tablet, etc. FIG. 1 may be an example of signaling and procedures for enabling fixed scalar or vector quantization in autoencoder-based CSI compression. FIG. 1 includes a one-sided joint training scheme where an entity (e.g., network 200) trains and shares encoder and decoder considering the quantization settings.

[0031] At S1010, the network may consider a quantization setting including distribution (DIST) for the levels / codewords of the scalar / vector quantizer, parameters (PARAM) related to the selected distribution, quantization metric (METRIC), scaling factor (SF) used to scale the quantization levels / codewords, segment size (SS), and bit per segment (BPS). The distribution is closely related to the encoder outputs. The parameters may include the range for a uniform distribution or the mean and variance for a Gaussian distribution. Also, in the case of vector quantization, the parameters may include correlation coefficient / matrix between the entries can be considered an additional parameter. The quantization metric is normally assumed to be Euclidean distance. However, a different metric may be chosen. For example, squared generalized cosine similarity may be chosen. The segment size may determine the dimensionality of the input of the quantization block. SS= I results in scalar quantization. SS greater than 1 may result in a vector quantization. BPS may determine how many bits are allocated for reporting values of a segment. The number of levels / codewords of the quantizer may also be determined by the BPS. For example, the number of levels / codewords may be 2 to the power of PBS (2BPS). Each of the DIST, PARAMS,Metric, SF, SS, and BPS, for an autoencoder and decoder may be determined and collectively may be referred to as fixed quantization settings.

[0032] Several quantization types can be considered for a quantization task, including uniform scalar quantization, non-uniform scalar quantization, and vector quantization. In scalar quantization (SQ), each encoder output entry (element) is quantized based on its own value and the quantization type and properties. However, in vector quantization (VQ), the encoder outputs are divided into some segments, and then the quantization is applied to each segment separately.

[0033] A distribution with uniformly spaced quantization levels leads to the construction of a universal scalar quantization (USQ). USQ is widely used as a fixed quantization scheme that can be used with quantization-aware training. Although USQ provides good performance in quantization-aware training, it may cause a constraint on the payload size, e.g., the feedback payload size with USQ may only be multiples of encoder output dimension. In a quantization-unaware training scheme USQ may not follow the distribution of the encoder outputs and may result in errors. Thus, for a quantization-unaware training scheme VQ may be utilized.

[0034] The network by considering the DIST, PARAMS, METRIC, SF, SS and BPS not only enables fixed vector quantization for quantization-aware training, but also enables fixed scalar / vector quantization for quantization-unaware training by allowing other codeword distributions rather than a uniform distribution. In addition, enabling fixed vector quantization allows more flexibility in CSI-feedback payload size.

[0035] Quantization levels and codewords (e.g., in a code book with each quantization level having at least one codeword) for each quantization level may be generated based on the determined fixed quantization settings (e.g., DIST, PARAMS, METRIC, SF, SS and BPS).

[0036] At S1020, the network 200 may train an autoencoder with the considered fixed quantization (e.g., with the fixed quantization settings determined at S1010) in a quantization aware or quantization un-aware scheme. The autoencoder may be trained using a deep learning machine learning algorithm (e.g., the autoencoder may be trained using a set of CSI using a deep learning machine learning algorithm or another machine learning algorithm). The decoder may likewise be prepared to perform the inverse of the encoding (i.e., decoding).

[0037] A transformer-based ML model (with 6 attention layers and about 8M parameters) may be used for both USQ and VQ training with the same number of overhead bits (128) in the simulations. For VQ- variational autoencoder (VAE), segment sizes of 1, 2, and 4 may be used, which results in 64, 32, and 16 segments and 2, 4, and 8 bits per segment, respectively. In case 1 of Table 1 for VQ is quantization-unaware training, and case 2 (including subcases 2-1 and 2-2) is quantization-aware training. For quantization-aware training, VQ-VAE performs slightly better than the USQ. In addition, if the fixed vector quantization codebook is well designed, quantization-unaware VQ can achieve similar performance to VQ-VAE.Table 1

[0038] Training an autoencoder considering a fixed vector quantization may be similar to the procedure in variational autoencoder architecture (VQ-VAE). Some differences are that the term in the loss function for updating the quantization codewords may be removed, as the quantization codewords may be fixed (not trainable). The loss function for training an autoencoder with fixed vector quantization can be defined as equation 1.(1) L = SGCS(H,H) + \\ze- e ||2

[0039] Where H, H, ze, and e denote, respectively, the original CSI, estimated CSI, compressed version of CSI, and quantization codewords.

[0040] At S1030, the network 200 may send an indication to enable fixed quantization to the UE 100 (e.g., “FixedQ=True”). The UE 100 may receive the indication to enable fixed quantization and enable fixed quantization or confirm that fixed quantization is enabled.

[0041] At S 1040, the UE 100 may send and the network 200 may receive a confirmation of the UE’ s capability for supporting fixed quantization or confirmation of enabled fixed quantization.

[0042] At S1050, the network 200 may share the selected settings for the fixed quantization with the UE 100. For example, the network 200 may send DIST, PARAMS, METRIC (if not Euclidean), SF, SS, and BPS. If a quantization metric other than Euclidean is chosen the quantization metric may be sent to the UE 100. If the METRIC is not sent it may be assumed by the UE 100 that Euclidean metrics are being used.

[0043] At S 1060, the UE 100 may construct the quantization levels and codewords based on the received settings for the fixed quantization. The quantization levels and codewords, may be constructed in the same way (or a similar way as) the network 200 constructs quantization levels and codewords using the fixed quantization settings. Restated, the UE100 may establish a quantizer with the quantization levels and codewords based on the received settings from the fixed quantization.

[0044] At S1070, the network 200 may communicate the encoder model structure and weights to the UE 100. The UE 100 may receive the encoder model structure and weights and construct an encoder substantially matching the encoder at the network 200 such that the decoder at network 200 can accurately decode that which is encoded by the encoder at the UE 100.

[0045] At S1080, the UE 100 may estimate the CSI for a channel which the UE 100 is connected to or communicating through. The UE 100 may estimate the channel state information using hardware on the UE 100 including processors, memory, clocks, transceivers, etc. based on communications through the channel.

[0046] At S1090, the UE 100 may compress the CSI using the encoder and quantize the compressed CSI using the quantizer. Methods of quantization are discussed in greater detail below.

[0047] At SHOO, the UE 100 may communicate the compressed quantized CSI to the network 200. The UE 100 may send and the network 200 may receive the communication.

[0048] At SI 110, the network may use the considered quantization settings for the fixed quantization to de-quantize the bits of the compressed quantized CSI. Methods of dequantization are discussed in greater detail below.

[0049] At SI 120, the network 200 may use the decoder to reconstruct (e.g., decompress) the CSI based on the de-quantized signal (e.g., de-quantized compressed CSI). Restated, the network 200 may decompress the compressed CSI using the decoder. Thus, the quantized compressed CSI can be reconstructed with high fidelity using the decoder and dequantizer of the network and overhead can be reduced for the channel for communicating CSI from the UE to the network 200.

[0050] Channel state information may be measured periodically by the UE 100 and shared with the network 200. Steps S1080-S1120 may be performed repeatedly to facilitate the sharing of the CSI.

[0051] FIG. 2 illustrates another example flow diagram 2000 of communications between the UE and the network element, according to certain example embodiments. Some of the operations of FIG. 2 may be substantially similar to the operations describe with relation to FIG. 1. Description of similar operations may be omitted or described in less detail.

[0052] FIG. 2 provides signaling for quantization alignment in a separate training scheme, where one of the entities (e.g., network) trains its encoder / decoder model and share a dataset with the other entity. Here, the considered fixed quantization is aligned based on the distribution (DIST) and parameters (PARAMS) at the first entity.

[0053] At S2010, the Network may consider a quantization setting including distribution (DIST) for the levels / codewords of the scalar / vector quantizer, PARAM, METRIC, SF, SS, and BPS.

[0054] At S2020, the network 200 may train an autoencoder with the considered fixed quantization in a quantization aware or quantization un-aware scheme. The autoencoder may be trained using a set of CSI and compressed CSI using a deep learning algorithm or another machine learning algorithm.

[0055] At S2030, the network 200 may send an indication to enable fixed quantization to the UE 100 (e.g., “FixedQ=True”). The UE 100 may receive the indication to enable fixed quantization and enable fixed quantization or confirm that fixed quantization is enabled.

[0056] At S2040, the UE 100 may send and the network 200 may receive a confirmation of the UE’ s capability for supporting fixed quantization or confirmation of enabled fixed quantization.

[0057] At S2050, the network 200 may share the selected settings for the fixed quantization with the UE 100. For example, the network 200 may send DIST, PARAMS, METRIC (if not Euclidean), SF, SS, and BPS.

[0058] At S2060, the UE 100 may construct the quantization levels and codewords based on the received settings for the fixed quantization.

[0059] At S2070, the network 200, may share the CSI training set (including original and compressed CSI) and the corresponding latent vectors according to the considered fixed quantization. Thus, the UE 100 may obtain the information needed to train the encoder at the UE 100 to encode in substantially the same way and the encoder at the network 200.

[0060] At S2080, the UE 100 may train the encoder based on the received dataset including the original CSI and compressed CSI. The encode at the UE 100 may be training using the same machine learning algorithm that the network trained the network encoder such that the decoder at the network 200 will accurately decode what the encoder at the UE 100 encodes.

[0061] At S2090, the UE 100 may estimate the CSI for a channel which the UE 100 is connected to or communicating through.

[0062] At S2100, the UE 100 may compress the CSI using the encoder and quantize the compressed CSI using the quantizer.

[0063] At S2110, the UE 100 may communicate the compressed quantized CSI to the network 200.

[0064] At S2120, the network may use the considered quantization settings for the fixed quantization to de-quantize the bits of the compressed quantized CSI.

[0065] At S2130, the network 200 may use the decoder to reconstruct (e.g., decompress) the CSI based on the de-quantized signal (e.g., de-quantized compressed CSI).

[0066] FIG. 3 illustrates another example flow diagram 3000 of communications between the user equipment 100 and the network 200, according to certain example embodiments. Some of the operations of FIG. 3 may be substantially similar to the operations describe with relation to FIGs. 1 and 2. Description of similar operations may be omitted or described in less detail.

[0067] FIG. 3 provides signaling for sharing quantization levels and codebook in fixed quantization. UE 100 or network 200 may share the quantization alphabet (levels or codebook) with the other entity. The overhead of this embodiment depends on the cardinality of the quantization codebook.

[0068] At S3010, the Network may consider a quantization setting including distribution (DIST) for the levels / codewords of the scalar / vector quantizer, PARAM, METRIC, SF, SS, and BPS.

[0069] At S3020, the network 200 may train an autoencoder with the considered fixed quantization in a quantization aware or quantization un-aware scheme. The autoencoder may be trained using a set of CSI using a deep learning machine learning algorithm or another machine learning algorithm.

[0070] At S3030, the network 200 may send an indication to enable fixed quantization to the UE 100 (e.g., “FixedQ=True”). The UE 100 may receive the indication to enable fixed quantization and enable fixed quantization or confirm that fixed quantization is enabled.

[0071] At S2040, the UE 100 may send and the network 200 may receive a confirmation of the UE’ s capability for supporting fixed quantization or confirmation of enabled fixed quantization.

[0072] At S3050, the network 200 may share the selected quantization levels, codewords and codebook(s) for the fixed quantization with the UE 100. For example, the network 200 may send a code book with at least one codeword for each level of the fixed quantization.

[0073] At S3060, the UE 100 may construct the quantizer block (or quantizer) based on the received levels, codewords, and codebook(s).

[0074] At S3070, the network 200, may share the CSI training set (including original and compressed CSI) and the corresponding latent vectors according to the considered fixed quantization. Thus the UE 100 may obtain the information needed to train the encoder at the UE 100 to encode in substantially the same way as the encoder at the network 200.

[0075] At S3080, the UE 100 may train the encoder based on the received dataset including the original CSI and compressed CSI. The encode at the UE 100 may be training using the same machine learning algorithm that the network trained the network encoder such that the decoder at the network 200 will accurately decode what the encoder at the UE 100 encodes.

[0076] At S3090, the UE 100 may estimate the CSI for a channel which the UE 100 is connected to or communicating through.

[0077] At S3100, the UE 100 may compress the CSI using the encoder and quantize the compressed CSI using the quantizer.

[0078] At S3110, the UE 100 may communicate the compressed quantized CSI to the network 200.

[0079] At S3120, the network may use the considered quantization settings for the fixed quantization to de-quantize the bits of the compressed quantized CSI.

[0080] At S3130, the network 200 may use the decoder to reconstruct (e.g., decompress) the CSI based on the de-quantized signal (e.g., de-quantized compressed CSI).

[0081] The progression of steps discussed in FIGs. 1-3 are not mutually exclusive and can be combined or modified. The operations may also not be exclusive to reporting CSI. Generally, a first entity (NW 200, UE 100, or other) may select a training scheme (quantization-aware / -unaware), quantization settings and parameters, encoder / decoder model architecture(s), and compression rate. The first entity may train the encoder / decoder according to the selected training scheme. The first entity shares the quantization settings, considered distribution for the quantization, and quantization parameters with a second entity (NW 200, UE 100 or other). Thus, the first entity and second entity share (e.g., by either entity sending to the other) the quantization parameters such that the either can quantize and dequantize information in a substantially similar way. The second entity may construct the quantization level s / c ode words based on the received quantization settings and parameters. The second entity (if needed) may train / fine-tune its encoder or decoder based on the received information and quantization settings from the first entity. The first and second entity may then communicate using the coordinated encoder / decoder and quantization / dequantization blocks .

[0082] Further, other operations may be performed such as handshake operations, and exchanges of indications of capabilities and permissions which may be considered when constructing / establishing the encoder / decoder and quantization / dequantization blocks.

[0083] FIG. 4 illustrates an example flow chart 4000 for a method of controlling a user equipment, according to certain example embodiments. In an example embodiment, the method of FIG. 4 may be performed by a network entity, or a group of multiple network elements in a 3GPP system, such as LTE or 5G-NR. For instance, in an exampleembodiment, the method of FIG. 4 may be performed by the network (e.g., gNB, 5G core, or 5G RAN) similar to one of apparatuses 10 or 20 illustrated in FIG. 8.

[0084] According to certain example embodiments, the method of FIG. 5 may include, at 45010, training, at a network device, at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme, at S4020, determining quantization block information, the quantization block information being based on the at least one fixed quantization setting, at S4030, sending, to a user equipment, encoder model information, at S4040, obtaining channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

[0085] FIG. 5 illustrates an example flow chart 5000 for a method of controlling a network element, according to certain example embodiments. In an example embodiment, the method of FIG. 5 may be performed by a network entity, or a group of multiple network elements in a 3GPP system, such as LTE or 5G-NR, 6G, 7G, etc. For instance, in an example embodiment, the method of FIG. 5 may be performed by a UE or computer implementing an application and / or AF, the UE and computer being similar to one of apparatuses 10 or 20 illustrated in FIG. 8.

[0086] According to certain example embodiments, the method of FIG. 5 may include, at S5010, establishing, at a user equipment, a quantization block based, at least partially, on quantization block information, at S5020, establishing, at the user equipment, an encoder based, at least partially, on received encoder model information from a network device, at S5030, estimating channel state information of a channel, at S5040, compressing the channel state information of the channel using the quantization block and the encoder, and, at S5050, sending, to the network device, the compressed channel state information.

[0087] FIG. 6A illustrates an example vector quantization, according to certain example embodiments. FIG. 6B illustrates another example vector quantization with 2D uniformdistribution, according to certain example embodiments. Both FIG. 6A and 6B show SS of 2. FIG. 6 A shows uniform distribution in the range [-1, 1] X[-l, 1] and BPS = 4 results in the 2A4= 16 codewords. FIG. 6B shows BPS=3, the corresponding 2A3=8 codewords. The grid of 16 code words shown in FIG. 6A can be obtained using USQ, however, the 8 codewords pattern of FIG. 6B cannot be build or constructed with USQ. Using the example vector quantization shown in FIGs. 6 A and 6B a value in each of the boxes along the 2 dimensional plane can be represented with a single code word. For example, using the lowermost leftmost block of FIG. 6A, the codeword for the lowermost leftmost block may be used to quantize any pair of values on the two dimensional plane that falls in the block (e.g., -.75, -.75). The point shown in the center of the block may be used as the values when dequantizing the codeword. This may be done with any set of values in the range [-1, 1] X [-1, 1]. These are only examples of quantization and dequantization for SS=2 other values of SS may also be used. The range may be scaled as needed by the value of SF (e.g., a range of [-2, 2] X[-2, 2] when a SF=2).

[0088] FIG. 7A illustrates an example vector quantization with 2D standard Gaussian distribution, according to certain example embodiments. FIG. 7B illustrates another example vector quantization with 2D standard Gaussian distribution, according to certain example embodiments. FIG. 7A shows vector quantization SS=2 with 2D standard Gaussian distribution including 16 codewords (BPS = 4) with scaling SF = 1 / 4. FIG. 7B shows vector quantization SS=2 with 2D standard Gaussian distribution including 16 codewords (BPS = 4) with scaling SF = 1 / 3. The regions for each quantization codeword are not shown, however, the values which each quantization codeword will be assigned when dequantized is shown, the regions are based on which of the dequantization values is closest to the pair of values being quantized.

[0089] The settings of the quantizer / dequantizer and the encoder / decoder may be considered and determined or set based on the CSI which is being compressed and quantized. For example, if the encoding of the CSI generates a pair of values, and the pairof values is more likely to be near [0,0], SS=2 with 2D standard Gaussian distribution including 16 codewords (BPS = 4) with scaling SF = 1 / 3 such as shown in FIG. 7A may be selected. If the encoded value has a more uniform distribution and is bounded by a range of [-1, 1] X [-1, 1] and it is preferred to reduce the number of bits to 3 for the codeword, then BPS=3 such as shown in FIG. 6B may be selected. Other considerations such as quantization degradation requirements or thresholds, quantization-aware or non-aware, overhead cost (e.g., bit limitations), may be considered as well when determining the fixed quantization settings.

[0090] To measure the degradation caused by the quantization block, we may define the quantization loss metric (MSEQ) as shown in equation 2.

[0091] Where Xeand X denote, respectively, the input and output signals of the quantization block. In SQ, Xeis a 1-dimensional signal (scalar). However, in VQ, Xeis an m-dimensional signal, where m denotes the segment size (codeword length). In other words, Xedenotes a segment of the encoder output Ze.

[0092] FIG. 8 illustrates a set of apparatuses 10 and 20. according to certain example embodiments. In certain example embodiments, apparatuses 10 and 20 may be elements in a communications network or associated with such a network. For example, apparatus 10 may be a UE, and apparatus 20 may be a network (e.g., gNB, 5GS, 5G Core, 5G RAN, etc.).

[0093] In some example embodiments, apparatuses 10 and 20 may include one or more processors, one or more computer-readable storage medium (for example, memory, storage, or the like), one or more radio access components (for example, a modem, a transceiver, or the like), and / or a user interface. In some example embodiments, apparatuses 10 and 20 may be configured to operate using one or more radio access technologies, such as GSM, LTE, LTE-A, NR, 5G, 6G, 7G, WLAN, WiFi, NB-IoT, Bluetooth, NFC, MulteFire, and / or anyother radio access technologies. It should be noted that one of ordinary skill in the art would understand that apparatuses 10 and 20 may include components or features not shown in FIG. 8.

[0094] As illustrated in the example of FIG. 8, apparatuses 10 and 20 may include or be coupled to a processors 12 and 22 for processing information and executing instructions or operations. Processors 12 and 22 may be any type of general or specific purpose processor. In fact, processors 12 and 22 may include one or more of general-purpose computers, special purpose computers, microprocessors, digital signal processors (DSPs), field-programmable gate arrays (FPGAs), application- specific integrated circuits (ASICs), and processors based on a multi-core processor architecture, as examples. While a single processor 12, 22 is shown in FIG. 8, multiple processors may be utilized according to other example embodiments. For example, it should be understood that, in certain example embodiments, apparatuses 10 and 20 may include two or more processors that may form a multiprocessor system (e.g., in this case processors 12 may represent a multiprocessor) that may support multiprocessing. According to certain example embodiments, the multiprocessor system may be tightly coupled or loosely coupled (e.g., to form a computer cluster).

[0095] Processors 12 and 22 may perform functions associated with the operation of apparatuses 10 and 20 including, as some examples, precoding of antenna gain / phase parameters, encoding and decoding of individual bits forming a communication message, formatting of information, and overall control of the apparatuses 10 and 20, including processes and examples illustrated in FIGs. 1-7B.

[0096] Apparatuses 10 and 20 may further include or be coupled to a memories 14 and 24 (internal or external), which may be respectively coupled to processors 12 and 24 for storing information and instructions that may be executed by processors 12 and 24. Memories 14 and 24 may be one or more memories and of any type suitable to the local application environment, and may be implemented using any suitable volatile or nonvolatile data storage technology such as a semiconductor-based memory device, a magnetic memorydevice and system, an optical memory device and system, fixed memory, and / or removable memory. For example, memories 14 and 24 can be comprised of any combination of random access memory (RAM), read only memory (ROM), static storage such as a magnetic or optical disk, hard disk drive (HDD), or any other type of non-transitory machine or computer readable media. The instructions stored in memories 14 and 24 may include program instructions or computer program code that, when executed by processors 12 and 22, enable the apparatuses 10 and 20 to perform tasks as described herein.

[0097] In certain example embodiments, apparatuses 10 and 20 may further include or be coupled to (internal or external) a drive or port that is configured to accept and read an external computer readable storage medium, such as an optical disc, USB drive, flash drive, or any other storage medium. For example, the external computer readable storage medium may store a computer program or software for execution by processors 12 and 22 and / or apparatuses 10 and 20 to perform any of the methods and examples illustrated in FIGs. 1- 7B.

[0098] In some example embodiments, apparatuses 10 and 20 may also include or be coupled to one or more antennas 15 and 25 for receiving a downlink signal and for transmitting via an UL from apparatuses 10 and 20. Apparatuses 10 and 20 may each further include a transceiver 18 and 28 configured to transmit and receive information. The transceivers 18 and 28 may also include a radio interface (e.g., a modem) coupled to the antennas 15 and 25. The radio interface may correspond to a plurality of radio access technologies including one or more of GSM, LTE, LTE-A, 5G, NR, WLAN, NB-IoT, Bluetooth, BT-LE, NFC, RFID, UWB, and the like. The radio interface may include other components, such as filters, converters (for example, digital-to-analog converters and the like), symbol demappers, signal shaping components, an Inverse Fast Fourier Transform (IFFT) module, and the like, to process symbols, such as OFDMA symbols, carried by a downlink or an UL.

[0099] For instance, transceivers 18 and 28 may be configured to modulate information on to a carrier waveform for transmission by the antennas 15 and 25 and demodulate information received via the antenna 15 and 25 for further processing by other elements of apparatuses 10 and 20. In other example embodiments, transceivers 18 and 28 may be capable of transmitting and receiving signals or data directly. Additionally or alternatively, in some example embodiments, apparatus 10 may include an input and / or output device (I / O device). In certain example embodiments, apparatuses 10 and 20 may further include a user interface, such as a graphical user interface or touchscreen.

[0100] In certain example embodiments, memories 14 and 34 store software modules that provide functionality when executed by processors 12 and 22. The modules may include, for example, an operating system that provides operating system functionality for apparatuses 10 and 20. The memory may also store one or more functional modules, such as an application or program, to provide additional functionality for apparatuses 10 and 20. The components of apparatuses 10 and 20 may be implemented in hardware, or as any suitable combination of hardware and software. According to certain example embodiments, apparatuses 10 and 20 may optionally be configured to communicate each other (in any combination) via a wireless or wired communication links 70 according to any radio access technology, such as NR.

[0101] According to certain example embodiments, processors 12 and 22 and memories 14 and 24 may be included in or may form a part of processing circuitry or control circuitry. In addition, in some example embodiments, transceivers 18 and 28 may be included in or may form a part of transceiver circuitry.

[0102] A computer program product may include one or more computer-executable components which, when the program is run, are configured to carry out some example embodiments. The one or more computer-executable components may be at least one software code or portions of it. Modifications and configurations required for implementing functionality of certain example embodiments may be performed as routine(s), which maybe implemented as added or updated software routine(s). Software routine(s) may be downloaded into the apparatus.

[0103] As an example, software or a computer program code or portions of it may be in a source code form, object code form, or in some intermediate form, and it may be stored in some sort of carrier, distribution medium, or computer readable medium, which may be any entity or device capable of carrying the program. Such carriers may include a record medium, computer memory, read-only memory, photoelectrical and / or electrical carrier signal, telecommunications signal, and software distribution package, for example. Depending on the processing power needed, the computer program may be executed in a single electronic digital computer or it may be distributed amongst a number of computers. The computer readable medium or computer readable storage medium may be a non- transitory medium.

[0104] In other example embodiments, the functionality may be performed by hardware or circuitry included in an apparatus (e.g., apparatus 10 or apparatus 20), for example through the use of an application specific integrated circuit (ASIC), a programmable gate array (PGA), a field programmable gate array (FPGA), or any other combination of hardware and software. In yet another example embodiment, the functionality may be implemented as a signal, a non-tangible means that can be carried by an electromagnetic signal downloaded from the Internet or other network.

[0105] According to certain example embodiments, an apparatus, such as a node, device, or a corresponding component, may be configured as circuitry, a computer or a microprocessor, such as single-chip computer element, or as a chipset, including at least a memory for providing storage capacity used for arithmetic operation and an operation processor for executing the arithmetic operation.

[0106] One having ordinary skill in the art will readily understand that the invention as discussed above may be practiced with procedures in a different order, and / or with hardwareelements in configurations which are different than those which are disclosed. Therefore, although the invention has been described based upon these example embodiments, it would be apparent to those of skill in the art that certain modifications, variations, and alternative constructions would be apparent, while remaining within the spirit and scope of example embodiments. Although the above embodiments refer to 5G NR and LTE technology, the above embodiments may also apply to any other present or future 3GPP technology, such as LTE-advanced, and / or fourth generation (4G) technology.Partial Glossary

[0107] 3GPP 3rd Generation Partnership Project

[0108] 5G 5th Generation

[0109] 5GCN 5G Core Network

[0110] BS Base Station

[0111] DL Downlink

[0112] eNB Enhanced Node B

[0113] gNB 5G or Next Generation NodeB

[0114] LTE Long Term Evolution

[0115] NR New Radio

[0116] UE User Equipment

[0117] CSI Channel state information

[0118] MIMO Multiple-input multiple-output

[0119] CR compression rate

[0120] CS compressive sensing

[0121] AE autoencoder

[0122] DL deep learning

[0123] NW network

[0124] 2D two dimensional

[0125] USQ uniform scalar quantization

[0126] VQ vector quantization

Claims

WE CLAIM:

1. A method comprising: training, at a network device, at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme; determining quantization block information, the quantization block information being based on the at least one fixed quantization setting; sending, to a user equipment, encoder model information; and obtaining channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

2. The method of claim 1, wherein the obtaining channel state information further comprises decoding the channel state information based on the dequantized received compressed channel state information.

3. The method of claim 1 or 2, further comprising: sending the quantization block information to the user equipment.

4. The method of claim 1 or 2, further comprising: receiving, from the user equipment, the quantization block information.

5. The method of any of claims 1-4, further comprising: sending, to the user equipment, an indication to indicate enabling of fixed quantization; and receiving, from the user equipment, a confirmation of capability to support the fixed quantization.

6. The method of any of claims 1-5, wherein the encoder model information comprises at least one of encoder model structure or encoder model weights.

7. The method of any of claims 1-5, wherein the encoder model information comprises training channel state information and corresponding latent vectors based on the at least one fixed quantization settings.

8. The method of any of claims 1-7, wherein the quantization block information comprises the fixed quantization settings.

9. The method any of claims 1-7, wherein the quantization block information comprises selected levels and codebooks for the at least one fixed quantization setting.

10. A method comprising: establishing, at a user equipment, a quantization block based, at least partially, on quantization block information; establishing, at the user equipment, an encoder based, at least partially, on received encoder model information from a network device; estimating channel state information of a channel; compressing the channel state information of the channel using the quantization block and the encoder; and sending, to the network device, the compressed channel state information.

11. The method of claim 10, further comprising: receiving, from the network device, an indication to enable fixed quantization; and sending, to the network device, a confirmation of capability to support the fixed quantization.

12. The method of claim 10 or 11, wherein the encoder model information comprises at least one of encoder model structure or encoder model weights.

13. The method of claim 10 or 11, wherein the encoder model information comprises training channel state information and corresponding latent vectors based on at least one fixed quantization setting.

14. The method any of claims 10-13, wherein the quantization block information comprises at least one fixed quantization setting.

15. The method any of claims 10-13, wherein the quantization block information comprises selected levels and codebooks for the at least one fixed quantization setting.

16. An apparatus, comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: train at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme; determine quantization block information, the quantization block information being based on the at least one fixed quantization setting; send, to a user equipment, encoder model information; and obtain channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

17. The apparatus of claim 16, wherein the obtaining channel state information comprises the at least one memory stores instructions that, when executed by the at leastone processor, cause the apparatus to decode the channel state information based on the dequantized received compressed channel state information.

18. The apparatus of claim 16 or 17, wherein the at least one memory stores instructions that, when executed by the at least one processor, cause the apparatus to: send the quantization block information to the user equipment.

19. The apparatus of claim 16 or 17, wherein the at least one memory stores instructions that, when executed by the at least one processor, cause the apparatus to: receive, from the user equipment, the quantization block information.

20. The apparatus of any of claims 16-19, wherein the at least one memory stores instructions that, when executed by the at least one processor, cause the apparatus to: send, to the user equipment, an indication to indicate enabling of fixed quantization; and receive, from the user equipment, a confirmation of capability to support the fixed quantization.

21. The apparatus of any of claims 16-20, wherein the encoder model information comprises at least one of encoder model structure or encoder model weights.

22. The apparatus of any of claims 16-20, wherein the encoder model information comprises training channel state information and corresponding latent vectors based on the at least one fixed quantization settings.

23. The apparatus of any of claims 16-22, wherein the quantization block information comprises the fixed quantization settings.

24. The apparatus any of claims 16-22, wherein the quantization block information comprises selected levels and codebooks for the at least one fixed quantization setting.

25. An apparatus, comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: establish a quantization block based, at least partially, on quantization block information; establish an encoder based, at least partially, on received encoder model information from a network device; estimate channel state information of a channel; compress the channel state information of the channel using the quantization block and the encoder; and send, to the network device, the compressed channel state information.

26. The apparatus of claim 25, wherein the at least one memory stores instructions that, when executed by the at least one processor, cause the apparatus to: receive, from the network device, an indication to enable fixed quantization; and send, to the network device, a confirmation of capability to support the fixed quantization.

27. The apparatus of claim 25 or 26, wherein the encoder model information comprises at least one of encoder model structure or encoder model weights.

28. The apparatus of claim 25 or 26, wherein the encoder model information comprises training channel state information and corresponding latent vectors based on at least one fixed quantization setting.

29. The apparatus any of claims 25-28, wherein the quantization block information comprises at least one fixed quantization setting.

30. The apparatus any of claims 25-28, wherein the quantization block information comprises selected levels and codebooks for the at least one fixed quantization setting.

31. An apparatus, comprising: means for training at least one autoencoder with at least one fixed quantization setting in a quantization-aware or quantization un-aware scheme; means for determining quantization block information, the quantization block information being based on the at least one fixed quantization setting; means for sending, to a user equipment, encoder model information; and means for obtaining channel state information by de-quantizing a received compressed channel state information using the at least one fixed quantization setting.

32. The apparatus of claim 31 , wherein the means for obtaining channel state information further comprises means for decoding the channel state information based on the dequantized received compressed channel state information.

33. The apparatus of claim 31 or 32, further comprising: means for sending the quantization block information to the user equipment.

34. The apparatus of claim 31 or 32, further comprising: means for receiving, from the user equipment, the quantization block information.

35. The apparatus of any of claims 31-34, further comprising:means for sending, to the user equipment, an indication to indicate enabling of fixed quantization; and means for receiving, from the user equipment, a confirmation of capability to support the fixed quantization.

36. The apparatus of any of claims 31-35, wherein the encoder model information comprises at least one of encoder model structure or encoder model weights.

37. The apparatus of any of claims 31-35, wherein the encoder model information comprises training channel state information and corresponding latent vectors based on the at least one fixed quantization settings.

38. The apparatus of any of claims 31-37, wherein the quantization block information comprises the fixed quantization settings.

39. The apparatus any of claims 31-37, wherein the quantization block information comprises selected levels and codebooks for the at least one fixed quantization setting.

40. An apparatus, comprising: means for establishing a quantization block based, at least partially, on quantization block information; means for establishing an encoder based, at least partially, on received encoder model information from a network device; means for estimating channel state information of a channel; means for compressing the channel state information of the channel using the quantization block and the encoder; and means for sending, to the network device, the compressed channel state information.

41. The apparatus of claim 40, further comprising: means for receiving, from the network device, an indication to enable fixed quantization; and means for sending, to the network device, a confirmation of capability to support the fixed quantization.

42. The apparatus of claim 40 or 41, wherein the encoder model information comprises at least one of encoder model structure or encoder model weights.

43. The apparatus of claim 40 or 41, wherein the encoder model information comprises training channel state information and corresponding latent vectors based on at least one fixed quantization setting.

44. The apparatus any of claims 40-43, wherein the quantization block information comprises at least one fixed quantization setting.

45. The apparatus any of claims 40-43, wherein the quantization block information comprises selected levels and codebooks for the at least one fixed quantization setting.

46. An apparatus, comprising: circuitry configured to perform the method according to any of claims 1-15.

47. A non-transitory computer readable medium comprising program instructions stored thereon for performing at least the method according to any of claims 1-15.

Images