Fig 2 - uploaded by Henri Ruotsalainen
Content may be subject to copyright.
Complete baseband model consisting of complex imbalance regeneration part and two layer time delay artificial neural network.
Source publication
In this work a novel time delay neural network based complex baseband model is proposed for fully digital pulse driven power amplifier concepts. It will be shown that the given technique is able to capture the non-linear effects related to the analog components of the transmitter. Results based on simulations and measurements indicate an accurate m...
Context in source publication
Context 1
... proposed model given in Fig. 2 consists of two blocks: complex imbalance regeneration and nISI TDNN. The first sub-block performs data preparation by converting the complex valued pulse sequence x(n) to include the complex imbalances related to the digital up-conversion. A more detailed descrip- tion of the first sub-block can be found in [8]. The second sub-block implements the actual nISI modeling by a two layer TDNN operating for real valued quadrature sequences. The first layer is the input layer composed of a delay line with memory depth of N m connected to variable number of neurons N i with sigmoid activation functions. The second layer is the output layer with two linear activation functions. Thus, a single activation function of the input layer receives 4N m sequences in total. Moreover, at least two input layer activation functions are required to enable correct mapping of quadrature sequences between the input and the output ...
Citations
... Microwave amplifiers add noise to the coveted signal delivering corruption of affectability, determination and signal quality in wireless systems. NF is usually decided without the carrier, yet it can likewise be acquired from AM or PM noise spectra and, along these lines, as an element of carrier level [4][5][6]. ...
In this paper, the test portrayal strategies of nonlinear measurements of wideband LNA exhibitions are simulated. Likewise, significance of the nonlinear estimations has been portrayed unmistakably with microwave LNA operating large signal analysis had been assessed. This work is endeavored to demonstrate the performances of the LNA with AM-AM, AM-PM measurements in points of interest and differentiating from the linear measurement regime. The most of the important aspects of LNA will be in linear measurements and furthermore to quantify nonlinear measurements accurately harmonic balance simulator is used. The harmonics up to order 3 and power characteristics are altogether shown with power swept variable. A simulation setup is made to measure the characteristics of LNA by using spectrum rectangular display type with power harmonic components. At last, author designed wideband LNA from the bandwidth 1 GHz to 5 GHz and elaborates how nonlinear measurements changed the way of LNA design to validate and development in microwave frequencies.
... Surprisingly, not much research has been done on behavioral modelling of all-digital transmitters. Neural networks based behavioral ADT models were presented in [8]. Main deficiency of such black box modeling is the lack of insight into structure and internal dynamics of the nonlinear system being approximated. ...
In this paper an exact baseband equivalent model of a quadrature modulated all-digital transmitter is derived. No restrictions on the number of levels of a digital switched-mode power amplifier (SMPA) driving input, nor the pulse encoding scheme employed, are made. This implies a high level of generality of the proposed model. We show that all-digital transmitter (ADT) can be represented as a series connection of the pulse encoder, discrete-time Volterra series model of fixed degree and memory depth, and a linear time-varying system with special properties. This result suggests a new analytically motivated structure of a digital predistortion (DPD) of SMPA nonlinearities in ADT. Numerical simulations in MATLAB are used to verify proposed baseband equivalent model.
... Such models, however, are able to capture only nISI stemming from the current and the previous RF bit symbol, which makes the approach less suitable for scenarios where strong memory effects are present. More advanced modeling techniques involve circuit domain nISI model [12], look-up table (LUT) model [20], Volterra based nISI model [21], time delay neural network (TDNN) based model [22] as well as non-linear autoregressive (NARX) model [23], which are more suitable to capture non-linear memory effects. The notable drawback of the advanced models operating directly on pulsed RF excitations, is firstly the exponentially growing computational complexity of Volterra based models. ...
... For example, the RF pulse encoding based on digital up-conversion of pulse encoded baseband sequences [5] leads to redundancy in the pulsed RF sequences. In our previous work [22], we showed that the redundancy can be exploited to perform the nISI modeling for the over-sampled, encoded complex baseband sequences instead for RF symbols. ...
... Moreover, due to lower modeling sampling rate requirements, larger memory depths in terms of RF symbols can be accessed. The main difference between the presented and the previous work of the authors is as follows: The new complex baseband representation is able to exactly represent the behavior of a Volterra based non-linear system, whereas the TDNN based model given in [22] can be considered as an approximative solution of the presented model. Furthermore, by introducing a band-limited nISI kernel related to the band-limited Volterra kernel [25], [16], the presented nISI model is capable of modeling the entire transmitter system, which is inherently band-limited due to the reconstruction filter. ...
In this paper an equivalent complex baseband representation of the analog component related non-linearity of digital transmitters relying on 1-bit complex baseband encoding is derived. By exploiting the properties of the pulsed RF encoding the novel behavioral modeling technique is able to represent accurately the non-linear memory effects of the power amplification stage. Furthermore, a band-limited kernel technique leads to more efficient modeling of the complete digital transmitter, and to relaxed sampling rate. For the parameter estimation the linear regression, common to Volterra model identification, can be employed. According to the simulation and measurement based verification results, the novel modeling technique excels the state-of-the-art in terms of modeling accuracy. It can be assumed that the given methodology serves both as a basis for future behavioral models and for the development of advanced encoding techniques for linearization purposes.
In this paper an exact baseband equivalent model of a quadrature modulated all-digital transmitter is derived. No restrictions on the number of levels of a digital switched-mode power amplifier (SMPA) driving input, nor the pulse encoding scheme employed, are made. This implies a high level of generality of the proposed model. We show that all-digital transmitter (ADT) can be represented as a series connection of the pulse encoder, discrete-time Volterra series model of fixed degree and memory depth, and a linear time-varying system with special properties. This result suggests a new analytically motivated structure of a digital predistortion (DPD) of SMPA nonlinearities in ADT. Numerical simulations in MATLAB are used to verify proposed baseband equivalent model.
In this paper, hybrid algorithm called gradient particle swarm optimization (GPSO) is proposed for training artificial neural networks (ANN). Then, the trained networks are applied to modeling waveguide filters (broad-band e-plane filters with improved stop-band and rectangular waveguide h-plane three-cavity filter). For validate effectiveness of this algorithm, we compared the results of convergence and modeling obtained with those obtained by back- propagation neural networks (BP-NN) and particle swarm optimization neural networks (PSO-NN).
Transmitters based on nonlinear radio-frequency (RF) modulators and switched-mode power amplifiers are systems that, at least theoretically, can provide high linearity and energy-efficient generation of RF signals at the same time. However, the nonlinear memory-affected behavior remains an issue hindering practical applications. This paper shows that it is possible to model the nonlinear memory of such circuits based on time-domain observations and how to use this information in computationally efficient time-domain RF circuit models. In contrast to other works, this model covers the full bandwidth of the active device (dc-10 GHz) and it uses hierarchical data structuring to adaptively find a compact model without prior knowledge of the circuit's memory depth. The data for this work were gained from a laboratory setup, designed to be used as an LTE transmitter for a signal bandwidth of 20 MHz at a fixed center frequency of 2.5 GHz.
Recent developments in the field of wireless communications emphasis the urgent need for flexible radio frequency (RF) front ends. Not only a multitude of frequency bands needs to be supported, also modulation formats come in a greater variety and tend to develop at a faster pace. While receiver topologies have been adapted by employing software defined radio (SDR) concepts, translating these developments to transmitters was hindered. Mostly due to the lack of power amplifiers that are energy efficient, linear and broadband at the same time. All digital radio frequency transmitters (DRFTx) are a promising, however, not a very mature option to overcome this issue. While offering the desired flexibility and potential energy efficient operation, these setups exhibit relatively long memory effects which also result in nonlinear inter symbol interference (nISI). In this paper it is demonstrated that the observed nonlinear memory effects, dominantly caused by the need for a reflective reconstruction filter, can be modeled reliably by a lookup table (LUT) model. This analysis is based on measured data of a DRFTx lab setup. However, they are backed up by simulations gained from readily available nonlinear transistor models within an RF computer aided design (CAD) tool.
In this paper a methodology for identifying broadband nonlinear models for digital switched mode RF power amplifiers is presented. This can be accomplished by training a black box model by observing an amplifier's output waveform in time domain. It is demonstrated that the presented method is feasible for using a simple and relatively short excitation sequence. The required model-complexity for a given amplifier is unknown in general, therefore, a testing method is presented in order to select a model of sufficient complexity. This is necessary since selecting a model solely on how well it reproduces the training sequence has been proven not to be sufficient. A digital power amplifier lab-setup serves as test platform in order to verify the validity of the so derived model.
