Sex differences in physiological correlates of affectively driven decision-making behavior in adult ADHD.

BACKGROUND: Sex differences in the symptomatology of adults with attention-deficit/hyperactivity disorder (ADHD) have often been overlooked when studying behavioral abnormalities. However, it is known that women exhibit considerably more stronger symptoms related to emotional competence than men. Since affective functions significantly influence the processing of risky decision-making and risk-engagement, we assume that risky behavior in ADHD is affected by sex differences. Therefore, we specifically investigated sex-specific effects on the interaction between emotionally induced changes in physiology and behavioral performance on a decision-making task. METHODS: Skin conductance responses of twenty-nine adults with ADHD (n = 16 male; n = 13 female) and thirty-three adults in the control group (n = 14 male; n = 19 female) were recorded during the performance in a modified version of the Balloon Analogue Risk Task (BART). Additional questionnaires were used to reveal insights in the self-assessment of emotional competence, risk perception, and feedback sensitivity. Emotional arousal and decision-making behavior were analyzed using linear mixed-effects models. RESULTS: Results showed different effects of sex on risk behaviors in controls and ADHD. In contrast to healthy controls, female adults with ADHD showed a significantly greater risk engagement in the BART compared to males with ADHD. This contrary sex relation was not observed in skin conductance responses and revealed a significantly different sex-dependent correlation of body response and behavioral task performance in ADHD. Comparisons with results from self-assessments furthermore indicate a reduced behavioral self-perception in women with ADHD, but not in men. CONCLUSION: In summary, we found an altered interaction between physiological activity and risky behavior in women with ADHD. Thus, the present study indicates a reduced sensitivity towards the own bodily responses in women with ADHD, which could consequently cause increased risky DM behavior in daily life. The current results suggest that more consideration needs to be given to sex-specific effects on physiological processes and behavior in adults with ADHD.

Multiscale heterogeneity of white matter morphometry in psychiatric disorders.

Background: Inter-individual variability in neurobiological and clinical characteristics in mental illness is often overlooked by classical group-mean case-control studies. Studies using normative modelling to infer person-specific deviations of grey matter volume have indicated that group means are not representative of most individuals. The extent to which this variability is present in white matter morphometry, which is integral to brain function, remains unclear. Methods: We applied Warped Bayesian Linear Regression normative models to T1-weighted magnetic resonance imaging data and mapped inter-individual variability in person-specific white matter volume deviations in 1,294 cases (58% male) diagnosed with one of six disorders (attention-deficit/hyperactivity, autism, bipolar, major depressive, obsessive-compulsive and schizophrenia) and 1,465 matched controls (54% male) recruited across 25 scan sites. We developed a framework to characterize deviation heterogeneity at multiple spatial scales, from individual voxels, through inter-regional connections, specific brain regions, and spatially extended brain networks. Results: The specific locations of white matter volume deviations were highly heterogeneous across participants, affecting the same voxel in fewer than 8% of individuals with the same diagnosis. For autism and schizophrenia, negative deviations (i.e., areas where volume is lower than normative expectations) aggregated into common tracts, regions and large-scale networks in up to 35% of individuals. Conclusions: The prevalence of white matter volume deviations was lower than previously observed in grey matter, and the specific location of these deviations was highly heterogeneous when considering voxel-wise spatial resolution. Evidence of aggregation within common pathways and networks was apparent in schizophrenia and autism but not other disorders.

Artificial Intelligence-Empowered Doppler Weather Profile for Low-Earth-Orbit Satellites.

Low-Earth-orbit (LEO) satellites are widely acknowledged as a promising infrastructure solution for global Internet of Things (IoT) services. However, the Doppler effect presents a significant challenge in the context of long-range (LoRa) modulation uplink connectivity. This study comprehensively examines the operational efficiency of LEO satellites concerning the Doppler weather effect, with state-of-the-art artificial intelligence techniques. Two LEO satellite constellations-Globalstar and the International Space Station (ISS)-were detected and tracked using ground radars in Perth and Brisbane, Australia, for 24 h starting 1 January 2024. The study involves modelling the constellation, calculating latency, and frequency offset and designing a hybrid Iterative Input Selection-Long Short-Term Memory Network (IIS-LSTM) integrated model to predict the Doppler weather profile for LEO satellites. The IIS algorithm selects relevant input variables for the model, while the LSTM algorithm learns and predicts patterns. This model is compared with Convolutional Neural Network and Extreme Gradient Boosting (XGBoost) models. The results show that the packet delivery rate is above 91% for the sensitive spread factor 12 with a bandwidth of 11.5 MHz for Globalstar and 145.8 MHz for ISS NAUKA. The carrier frequency for ISS orbiting at 402.3 km is 631 MHz and 500 MHz for Globalstar at 1414 km altitude, aiding in combating packet losses. The ISS-LSTM model achieved an accuracy of 97.51% and a loss of 1.17% with signal-to-noise ratios (SNRs) ranging from 0-30 dB. The XGB model has the fastest testing time, attaining ≈0.0997 s for higher SNRs and an accuracy of 87%. However, in lower SNR, it proves to be computationally expensive. IIS-LSTM attains a better computation time for lower SNRs at ≈0.4651 s, followed by XGB at ≈0.5990 and CNN at ≈0.6120 s. The study calls for further research on LoRa Doppler analysis, considering atmospheric attenuation, and relevant space parameters for future work.

Brain-age prediction: Systematic evaluation of site effects, and sample age range and size.

Structural neuroimaging data have been used to compute an estimate of the biological age of the brain (brain-age) which has been associated with other biologically and behaviorally meaningful measures of brain development and aging. The ongoing research interest in brain-age has highlighted the need for robust and publicly available brain-age models pre-trained on data from large samples of healthy individuals. To address this need we have previously released a developmental brain-age model. Here we expand this work to develop, empirically validate, and disseminate a pre-trained brain-age model to cover most of the human lifespan. To achieve this, we selected the best-performing model after systematically examining the impact of seven site harmonization strategies, age range, and sample size on brain-age prediction in a discovery sample of brain morphometric measures from 35,683 healthy individuals (age range: 5-90 years; 53.59% female). The pre-trained models were tested for cross-dataset generalizability in an independent sample comprising 2101 healthy individuals (age range: 8-80 years; 55.35% female) and for longitudinal consistency in a further sample comprising 377 healthy individuals (age range: 9-25 years; 49.87% female). This empirical examination yielded the following findings: (1) the accuracy of age prediction from morphometry data was higher when no site harmonization was applied; (2) dividing the discovery sample into two age-bins (5-40 and 40-90 years) provided a better balance between model accuracy and explained age variance than other alternatives; (3) model accuracy for brain-age prediction plateaued at a sample size exceeding 1600 participants. These findings have been incorporated into CentileBrain (https://centilebrain.org/#/brainAGE2), an open-science, web-based platform for individualized neuroimaging metrics.

Deep Learning Based Over-the-Air Training of Wireless Communication Systems without Feedback.

In trainable wireless communications systems, the use of deep learning for over-the-air training aims to address the discontinuity in backpropagation learning caused by the channel environment. The primary methods supporting this learning procedure either directly approximate the backpropagation gradients using techniques derived from reinforcement learning, or explicitly model the channel environment by training a generative channel model. In both cases, over-the-air training of transmitter and receiver requires a feedback channel to sound the channel environment and obtain measurements of the learning objective. The use of continuous feedback not only demands extra system resources but also makes the training process more susceptible to adversarial attacks. Conversely, opting for a feedback-free approach to train the models over the forward link, exclusively on the receiver side, could pose challenges to reliably end the training process without intermittent testing over the actual channel environment. In this article, we propose a novel method for the over-the-air training of wireless communication systems that does not require a feedback channel to train the transmitter and receiver. Random samples are transmitted through the channel environment to train a mixture density network to approximate the channel distribution on the receiver side of the network. The transmitter and receiver models are trained with the resulting channel model, and the transmitter can be deployed after training. We show that the block error rate measurements obtained with the simulated channel are suitable for monitoring as a stopping criterion during the training process. The resulting method is demonstrated to have equivalent performance to the end-to-end autoencoder training on small message sequences.

Practices of falls risk assessment and prevention in acute hospital settings: a realist investigation.

Background: Falls are the most common safety incident reported by acute hospitals. The National Institute of Health and Care Excellence recommends multifactorial falls risk assessment and tailored interventions, but implementation is variable. Aim: To determine how and in what contexts multifactorial falls risk assessment and tailored interventions are used in acute National Health Service hospitals in England. Design: Realist review and multisite case study. (1) Systematic searches to identify stakeholders' theories, tested using empirical data from primary studies. Review of falls prevention policies of acute Trusts. (2) Theory testing and refinement through observation, staff interviews (n = 50), patient and carer interviews (n = 31) and record review (n = 60). Setting: Three Trusts, one orthopaedic and one older person ward in each. Results: Seventy-eight studies were used for theory construction and 50 for theory testing. Four theories were explored. (1) Leadership: wards had falls link practitioners but authority to allocate resources for falls prevention resided with senior nurses. (2) Shared responsibility: a key falls prevention strategy was patient supervision. This fell to nursing staff, constraining the extent to which responsibility for falls prevention could be shared. (3) Facilitation: assessments were consistently documented but workload pressures could reduce this to a tick-box exercise. Assessment items varied. While individual patient risk factors were identified, patients were categorised as high or low risk to determine who should receive supervision. (4) Patient participation: nursing staff lacked time to explain to patients their falls risks or how to prevent themselves from falling, although other staff could do so. Sensitive communication could prevent patients taking actions that increase their risk of falling. Limitations: Within the realist review, we completed synthesis for only two theories. We could not access patient records before observations, preventing assessment of whether care plans were enacted. Conclusions: (1) Leadership: There should be a clear distinction between senior nurses' roles and falls link practitioners in relation to falls prevention; (2) shared responsibility: Trusts should consider how processes and systems, including the electronic health record, can be revised to better support a multidisciplinary approach, and alternatives to patient supervision should be considered; (3) facilitation: Trusts should consider how to reduce documentation burden and avoid tick-box responses, and ensure items included in the falls risk assessment tools align with guidance. Falls risk assessment tools and falls care plans should be presented as tools to support practice, rather than something to be audited; (4) patient participation: Trusts should consider how they can ensure patients receive individualised information about risks and preventing falls and provide staff with guidance on brief but sensitive ways to talk with patients to reduce the likelihood of actions that increase their risk of falling. Future work: (1) Development and evaluation of interventions to support multidisciplinary teams to undertake, and involve patients in, multifactorial falls risk assessment and selection and delivery of tailored interventions; (2) mixed method and economic evaluations of patient supervision; (3) evaluation of engagement support workers, volunteers and/or carers to support falls prevention. Research should include those with cognitive impairment and patients who do not speak English. Study registration: This study is registered as PROSPERO CRD42020184458. Funding: This award was funded by the National Institute for Health and Care Research (NIHR) Health and Social Care Delivery Research programme (NIHR award ref: NIHR129488) and is published in full in Health and Social Care Delivery Research; Vol. 12, No. 5. See the NIHR Funding and Awards website for further award information.

Many accidental falls by older people in hospitals could be avoided. There are guidelines to prevent falls, but some hospitals are better at following them than others. This study aimed to find out why. First, we looked at research and hospitals' falls policies for ideas about what stops falls. With advice from service users, we tested these ideas in four hospitals in England, watching how falls were prevented on wards for older people and people who need bone care, and talking to 50 staff, 28 patients and 3 carers. We found the following: Falls leadership: wards had staff called falls link practitioners who supported falls prevention, but senior nurses, not link practitioners, made the most important decisions. Sharing responsibility: patients with falls risks were monitored to try to stop falls. Because only nursing teams were always present to monitor patients, they had most responsibility for preventing falls. This limited sharing responsibility with other staff. Computer tools: nurses used computers to record prevention work, but high workloads could make this a 'tick-box' exercise. Computer tools reminded them to do this, although tools varied. Patients had individual falls plans, but they were also ranked more generally as high or low risk of falling, with 'high-risk' patients being monitored. Patient involvement: nursing staff did not have time to explain to patients how to prevent falls, but other staff could have such conversations. Many patients had problems like dementia and found it difficult to follow safety advice, although some could take steps to keep safe, with sensitive staff support. We need to involve patients, carers and different staff in falls prevention. Hospitals could develop computer systems to support this, think how to involve more ward staff, and provide guidance on helpful ways to talk with patients about falls.

An analysis of real-time suicidal ideation and its relationship with retrospective reports among young people with borderline personality disorder.

INTRODUCTION: This study aimed to analyze the real-time variability of suicidal ideation intensity and the relationship between real-time and retrospective reports of suicidal ideation made on the Beck Scale for Suicidal Ideation (BSS), among young people with borderline personality disorder (BPD). METHODS: Young people (15-25-year olds) with BPD (N = 46), recruited from two government-funded mental health services, rated the intensity of their suicidal ideation six times per day for 7 days before completing the BSS. RESULTS: For 70% of participants, suicidal ideation changed in intensity approximately five times across the week, both within and between days. BSS ratings were most highly correlated with the highest real-time ratings of suicidal ideation. However, this was not significantly different from the relationship between the BSS and both the average and most recent ratings. Median ratings of suicidal ideation intensity were higher on the BSS compared with an equivalent question asked in real time. CONCLUSION: Findings suggest that young people with BPD experience high levels of fluctuation in their intensity of suicidal ideation across a week and that retrospective reports of suicidal ideation might be more reflective of the most intense experience of suicidal ideation across the week.

Transdiagnostic risk identification: A validation study of the Clinical High At Risk Mental State (CHARMS) criteria.

A set of clinical criteria, the Clinical High At-Risk Mental State (CHARMS) criteria, have been developed to identify symptomatic young people who are at-risk of disorder progression. The current study aimed to validate the CHARMS criteria by testing whether they prospectively identify individuals at-risk of progressing from attenuated symptomatology to a first episode of serious mental disorder, namely first episode psychosis, first episode mania, severe major depression, and borderline personality disorder. 121 young people completed clinical evaluations at baseline, 6- and 12-month follow-up. The Kaplan-Meier method was used to assess transition rates. Cox regression and LASSO were used to examine baseline clinical predictors of transition. Linear mixed effects modelling was used to examine symptom severity. 28 % of CHARMS+ individuals transitioned to a Stage 2 disorder by 12-month follow-up. The CHARMS+ group had more severe symptoms at follow-up than the CHARMS- group. 96 % of Stage 2 transitions were initially to severe depression. Meeting criteria for multiple CHARMS subgroups was associated with higher transition risk: meeting one at-risk group = 24 %; meeting two at-risk groups = 17 %, meeting three at-risk groups = 55 %, meeting four at-risk groups = 50 %. The strongest baseline predictor of transition was severity of depressive symptoms. The CHARMS criteria identified a group of individuals at-risk of imminent onset of severe mental disorder, particularly severe depression. Larger scale studies and longer follow-up periods are required to validate and extend these findings.

Altered task-related decoupling of the rostral anterior cingulate cortex in depression.

Dysfunctional activity of the rostral anterior cingulate cortex (rACC) - an extensively connected hub region of the default mode network - has been broadly linked to cognitive and affective impairments in depression. However, the nature of aberrant task-related rACC suppression in depression is incompletely understood. In this study, we sought to characterize functional connectivity of rACC activity suppression ('deactivation') - an essential feature of rACC function - during external task engagement in depression. Specifically, we aimed to explore neural patterns of functional decoupling and coupling with the rACC during its task-driven suppression. We enrolled 81 15- to 25-year-old young people with moderate-to-severe major depressive disorder (MDD) before they commenced a 12-week clinical trial that assessed the effectiveness of cognitive behavioral therapy plus either fluoxetine or placebo. Ninety-four matched healthy controls were also recruited. Participants completed a functional magnetic resonance imaging face matching task known to elicit rACC suppression. To identify brain regions associated with the rACC during its task-driven suppression, we employed a seed-based functional connectivity analysis. We found MDD participants, compared to controls, showed significantly reduced 'decoupling' of the rACC with extended task-specific regions during task performance. Specifically, less decoupling was observed in the occipital and fusiform gyrus, dorsal ACC, medial prefrontal cortex, cuneus, amygdala, thalamus, and hippocampus. Notably, impaired decoupling was apparent in participants who did not remit to treatment, but not treatment remitters. Further, we found MDD participants showed significant increased coupling with the anterior insula cortex during task engagement. Our findings indicate that aberrant task-related rACC suppression is associated with disruptions in adaptive neural communication and dynamic switching between internal and external cognitive modes that may underpin maladaptive cognitions and biased emotional processing in depression.

Major depressive disorder associated alterations in the effective connectivity of the face processing network: a systematic review.

Major depressive disorder (MDD) is marked by altered processing of emotional stimuli, including facial expressions. Recent neuroimaging research has attempted to investigate how these stimuli alter the directional interactions between brain regions in those with MDD; however, methodological heterogeneity has made identifying consistent effects difficult. To address this, we systematically examined studies investigating MDD-associated differences present in effective connectivity during the processing of emotional facial expressions. We searched five databases: PsycINFO, EMBASE, PubMed, Scopus, and Web of Science, using a preregistered protocol (registration number: CRD42021271586). Of the 510 unique studies screened, 17 met our inclusion criteria. These studies identified that compared with healthy controls, participants with MDD demonstrated (1) reduced connectivity from the dorsolateral prefrontal cortex to the amygdala during the processing of negatively valenced expressions, and (2) increased inhibitory connectivity from the ventromedial prefrontal cortex to amygdala during the processing of happy facial expressions. Most studies investigating the amygdala and anterior cingulate cortex noted differences in their connectivity; however, the precise nature of these differences was inconsistent between studies. As such, commonalities observed across neuroimaging modalities warrant careful investigation to determine the specificity of these effects to particular subregions and emotional expressions. Future research examining longitudinal connectivity changes associated with treatment response may provide important insights into mechanisms underpinning therapeutic interventions, thus enabling more targeted treatment strategies.

Multi-site benchmark classification of major depressive disorder using machine learning on cortical and subcortical measures.

Machine learning (ML) techniques have gained popularity in the neuroimaging field due to their potential for classifying neuropsychiatric disorders. However, the diagnostic predictive power of the existing algorithms has been limited by small sample sizes, lack of representativeness, data leakage, and/or overfitting. Here, we overcome these limitations with the largest multi-site sample size to date (N = 5365) to provide a generalizable ML classification benchmark of major depressive disorder (MDD) using shallow linear and non-linear models. Leveraging brain measures from standardized ENIGMA analysis pipelines in FreeSurfer, we were able to classify MDD versus healthy controls (HC) with a balanced accuracy of around 62%. But after harmonizing the data, e.g., using ComBat, the balanced accuracy dropped to approximately 52%. Accuracy results close to random chance levels were also observed in stratified groups according to age of onset, antidepressant use, number of episodes and sex. Future studies incorporating higher dimensional brain imaging/phenotype features, and/or using more advanced machine and deep learning methods may yield more encouraging prospects.

Therapist competence, homework engagement, and client characteristics in CBT for youth depression: A study of mediation and moderation in a community-based trial.

OBJECTIVE: Prior studies of Cognitive Behavioral Therapy (CBT) have focused on the quantity and quality of clients' homework completion and only rarely have considered the role of therapist competence. METHODS: The present study examined (a) therapist competence across the entire process of integrating homework into CBT, including the review, design, and planning of tasks; (b) homework engagement, including client appraisals of the difficulty and obstacles encountered in task completion using the Homework Rating Scale - Revised (HRS-II); (c) pre-post symptom reduction as the index of outcome; and (d) considered client factors such as suicide risk in a community-based trial for adolescent depression. Trained independent observers assessed therapist competence and engagement with homework at two consecutive sessions of CBT for N = 80 young people (Mage = 19.61, SD = 2.60). RESULTS: Significant complementary mediation effects were obtained; there was an indirect mediation effect of HRS-II Beliefs (b = 1.03, SE B = 0.42, 95% BCa CI [0.35, 2.03]) and HRS-II Perceived Consequences on the Competence-Engagement relationship (b = 0.85, SE B = 0.31, 95% BCa CI [0.39, 1.61]). High levels of suicidal ideation were also shown to moderate this relationship. CONCLUSIONS: The present findings contribute to the growing body of CBT process research designed to examine the complex interrelationships of client and therapist variables, in a manner that reflects the actual process of therapy, and advances beyond studies of isolated predictors of symptom change.

Ensuring the affordable becomes accessible-lessons from ketamine, a new treatment for severe depression.

In this paper, the case study of ketamine as a new treatment for severe depression is used to outline the challenges of repurposing established medicines and we suggest potential solutions. The antidepressant effects of generic racemic ketamine were identified over 20 years ago, but there were insufficient incentives for commercial entities to pursue its registration, or support for non-commercial entities to fill this gap. As a result, the evaluation of generic ketamine was delayed, piecemeal, uncoordinated, and insufficient to gain approval. Meanwhile, substantial commercial investment enabled the widespread registration of a patented, intranasal s-enantiomeric ketamine formulation (Spravato®) for depression. However, Spravato is priced at $600-$900/dose compared to ~$5/dose for generic ketamine, and the ~AUD$100 million annual government investment requested in Australia (to cover drug costs alone) has been rejected twice, leaving this treatment largely inaccessible for Australian patients 2 years after Therapeutic Goods Administration approval. Moreover, emerging evidence indicates that generic racemic ketamine is at least as effective as Spravato, but no comparative trials were required for regulatory approval and have not been conducted. Without action, this story will repeat regularly in the next decade with a new wave of psychedelic-assisted psychotherapy treatments, for which the original off-patent molecules could be available at low-cost and reduce the overall cost of treatment. Several systemic reforms are required to ensure that affordable, effective options become accessible; these include commercial incentives, public and public-private funding schemes, reduced regulatory barriers and more coordinated international public funding schemes to support translational research.

The Addition of Fish Oil to Cognitive Behavioral Case Management for Youth Depression: A Randomized, Double-Blind, Placebo-Controlled, Multicenter Clinical Trial.

BACKGROUND: Clinical trials suggest that long-chain omega-3 polyunsaturated fatty acids (n-3 PUFAs) (fish oil) may reduce depressive symptoms in adults with major depressive disorder. Therefore, n-3 PUFAs may be a potential treatment for depression in youth. METHODS: Participants were 15- to-25 year-old individuals with major depressive disorder who sought care in one of three government-funded mental health services for young people in metropolitan Melbourne, Perth, or Sydney, Australia. Participants were randomly assigned in a double-blind, parallel-arm design to receive either fish oil (840 mg of eicosapentaenoic acid and 560 mg of docosahexaenoic acid) or placebo capsules as adjunct to cognitive behavioral case management. All participants were offered 50-minute cognitive behavioral case management sessions every 2 weeks delivered by qualified therapists (treatment as usual) at the study sites during the intervention period. The primary outcome was change in the interviewer-rated Quick Inventory of Depressive Symptomatology, Adolescent Version, score at 12 weeks. Erythrocyte n-3 PUFA levels were assessed pre-post intervention. RESULTS: A total of 233 young people were randomized to the treatment arms: 115 participants to the n-3 PUFA group and 118 to the placebo group. Mean change from baseline in the Quick Inventory of Depressive Symptomatology score was -5.8 in the n-3 PUFA group and -5.6 in the placebo group (mean difference, 0.2; 95% CI, -1.1 to 1.5; p = .75). Erythrocyte PUFA levels were not associated with depression severity at any time point. The incidence and severity of adverse events were similar in the two groups. CONCLUSIONS: This placebo-controlled trial and biomarker analysis found no evidence to support the use of fish oil for treatment in young people with major depressive disorder.

The self, neuroscience and psychosis study: Testing a neurophenomenological model of the onset of psychosis.

AIM: Basic self disturbance is a putative core vulnerability marker of schizophrenia spectrum disorders. The primary aims of the Self, Neuroscience and Psychosis (SNAP) study are to: (1) empirically test a previously described neurophenomenological self-disturbance model of psychosis by examining the relationship between specific clinical, neurocognitive, and neurophysiological variables in UHR patients, and (2) develop a prediction model using these neurophenomenological disturbances for persistence or deterioration of UHR symptoms at 12-month follow-up. METHODS: SNAP is a longitudinal observational study. Participants include 400 UHR individuals, 100 clinical controls with no attenuated psychotic symptoms, and 50 healthy controls. All participants complete baseline clinical and neurocognitive assessments and electroencephalography. The UHR sample are followed up for a total of 24 months, with clinical assessment completed every 6 months. RESULTS: This paper presents the protocol of the SNAP study, including background rationale, aims and hypotheses, design, and assessment procedures. CONCLUSIONS: The SNAP study will test whether neurophenomenological disturbances associated with basic self-disturbance predict persistence or intensification of UHR symptomatology over a 2-year follow up period, and how specific these disturbances are to a clinical population with attenuated psychotic symptoms. This may ultimately inform clinical care and pathoaetiological models of psychosis.

Channel-Agnostic Training of Transmitter and Receiver for Wireless Communications.

Wireless communications systems are traditionally designed by independently optimising signal processing functions based on a mathematical model. Deep learning-enabled communications have demonstrated end-to-end design by jointly optimising all components with respect to the communications environment. In the end-to-end approach, an assumed channel model is necessary to support training of the transmitter and receiver. This limitation has motivated recent work on over-the-air training to explore disjoint training for the transmitter and receiver without an assumed channel. These methods approximate the channel through a generative adversarial model or perform gradient approximation through reinforcement learning or similar methods. However, the generative adversarial model adds complexity by requiring an additional discriminator during training, while reinforcement learning methods require multiple forward passes to approximate the gradient and are sensitive to high variance in the error signal. A third, collaborative agent-based approach relies on an echo protocol to conduct training without channel assumptions. However, the coordination between agents increases the complexity and channel usage during training. In this article, we propose a simpler approach for disjoint training in which a local receiver model approximates the remote receiver model and is used to train the local transmitter. This simplified approach performs well under several different channel conditions, has equivalent performance to end-to-end training, and is well suited to adaptation to changing channel environments.