Claims:We Claim:

1. A system of quantum random number generator comprising:
a photon generator and detection module for photon generation, and detection;
a single-photon counting module, connected to the photon generator and detection module, for time tagging the detected photon; and
a processing module, connected to the single-photon counting module, for eliminating non-random component from the time tagged signal received from the single-photon counting module, and subsequently filtering non-obvious patterns from the signal to extract true quantum random numbers.

2. The system as claimed in claim 1, wherein the photon generator and detection module comprises:
a laser source for generating photons;
an optical attenuator for lowering the intensity of laser beam to a single-photon level;
at least one linear polarizer for polarizing incoming photons from the attenuated laser source;
at least one polarizing beam splitter for randomly transmitting or reflecting photons; and
at least one detector for detecting single-photon.

3. The system as claimed in claim 2, wherein the laser source includes but not limited to, a semiconductor laser diode, and the detector is a low efficiency single-photon detector such as LED (light emitting diode) operating in reverse bias.

4. The system as claimed in claim 1, wherein the processing module comprises:
an elimination unit configured to identify and eliminate non-random component including trend and seasonality from the signal received from the single-photon counting module;
a signal processing unit configured to process the residual signal received from the elimination unit, wherein the modulus of residual signal is computed and the signal is rounded to 0 or 1;
a machine learning unit configured to receive the processed signal from the signal processing unit and predict the signal having non-obvious pattern using neural network model; and
a processing unit, connected to the signal processing unit and the machine learning unit, for filtering the non-obvious pattern from the signal received from the signal processing unit and extracting true quantum random number, wherein the processing unit is configured to:
a) receive the signal from the signal processing unit and compare it with the predicted signal received from the machine learning unit to filter non-obvious pattern, wherein the part of signal which overlaps with each other are discarded to obtain true single-photon signal;
b) extract the data from the filtered signal in form of 0 and 1; and
c) store the data (true quantum random numbers) in a storage unit of the processing unit.

5. A method of quantum random number generator comprising the steps of:
a) generating and detecting a photon, by a photon generator and detection module;
b) time tagging the photon, by a single-photon counting module; and
c) eliminating non-random component from the signal received from the single-photon counting module, and subsequently filtering non-obvious pattern, through a processing module, to extract true quantum random numbers.

6. The method as claimed in claim 5, wherein the step of generating and detecting the photon includes the step of:
a) generating a photon by a laser source;
b) lowering the intensity of laser beam to the single-photon level, by an optical attenuator;
c) polarizing incoming photons, by a linear polarizer, from the attenuated laser source;
d) randomly transmitting or reflecting photons, by a polarizing beam splitter; and
e) detecting a single-photon, by a detector.

7. The method as claimed in Claim 5, wherein the step (c) of eliminating the non-random component from the signal received from the single-photon counting module, and subsequently filtering the non-obvious pattern, through a processing module, to extract true quantum random numbers, includes the steps of:
a) identifying and eliminating, by an elimination unit, non-random component, including trend and seasonality, from the signal received from the single-photon counting module;
b) processing residual signal received from the elimination unit, by a signal processing unit, wherein the modulus of residual signal is computed and the signal is rounded to 0 or 1;
c) predicting the signal having non-obvious pattern, by a machine learning unit, using a neural network model; and
d) processing, by a processing unit, to filter non-obvious pattern from the signal received from the signal processing unit and extract true quantum random number, wherein the processing by the processing unit comprises the steps of:
i. receiving the signal from the signal processing unit and comparing it with the predicted signal received from the neural network model of the machine learning unit to filter non-obvious pattern, wherein the part of signal which overlaps with each other are discarded to obtain true single-photon signal;
ii. extracting the data from the filtered signal in form of 0 and 1; and
iii. storing the data (true quantum random number) in a storage unit of the processing unit.

8. The method as claimed in claim 7, wherein the machine learning unit comprises the neural network model for predicting the signal to filter non-obvious pattern, the neural network model is developed using the steps:
a) preparing train and test data, wherein the signal received to the machine learning unit is split into train and test data sets in a specified ratio;
b) choosing a certain time window (w), wherein the time window (w) is treated as a parameter, the value of w for which the neural network model produces the best result is chosen;
c) preparing the data, wherein the train data is stacked into one dimensional arrays of size w so that it can be fed into the neural network model;
d) defining the neural network model, wherein neural network model consists of an input layer, a hidden layer, and an output layer; and
e) finalizing the neural network model, wherein the neural network model which has minimum error is finalized, an appropriate error metric is chosen to calculate the error between the original signal and the predicted signal by the NN post training.

9. The method as claimed in claim 5 or claim 7, wherein the step of filtering non-obvious pattern from the signal includes, but is not limited to, thermal noise, dark counts (false positives) in the detector(s), and jitters in the optical circuit due to spurious mechanical vibrations.

10. A method of quantum random number generator comprising the steps of:
a) eliminating non-random component, by a processing module, from a signal; and
b) filtering non-obvious pattern from the signal, by the processing module, to extract true quantum random numbers. , Description:FORM 2
THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENTS RULES, 2003

COMPLETE SPECIFICATION
(See section 10 and rule 13)

A SYSTEM AND METHOD OF QUANTUM RANDOM NUMBER GENERATOR

DEPASOLUTIONS PRIVATE LIMITED, a company incorporated under the laws of India, of the address No. 21, 1st B Main Road, Caveri Layout, Nagarbhavi Main Road, Vijayanagar, Bengaluru, Karnataka, Pincode – 560040, India.

The following specification particularly describes the invention and the manner in which it is to be performed.
FIELD OF THE INVENTION
The present invention relates to a system and method of quantum random number generator.

BACKGROUND OF THE INVENTION
Modern industry is highly data driven, and competitive edge depends on ensuring that the critical data is encrypted and only authorized users have access to the data. Random numbers have played a crucial role in solutions to provide secure communications to authenticate users including e-commerce, and financial transactions.

In addition, industries, such as, gaming industry, particularly the gambling sector, is highly dependent on good random number generator. The ability to produce not just good quality random numbers, but affordable truly random numbers benefits many industries.

Further, ensuring confidentiality during communication or transactions has been majorly dependent on using random numbers generated by mathematical algorithms. However, these are pseudo-random numbers, in that the sequence repeats itself after a period of time. Random numbers generated using quantum techniques are true random numbers, however, considering the high cost involved in generating random numbers using apparatus that exploit quantum properties, most organizations prefer to use mathematical algorithms.

Therefore, the object of the present invention is to solve one or more of aforementioned issues.

SUMMARY OF THE INVENTION
A system of quantum random number generator is described. The system of quantum random number generator comprising a photon generator and detection module for photon generation, and detection; a single-photon counting module, connected to the photon generator and detection module, for time tagging the detected photon; and a processing module, connected to the single-photon counting module, for eliminating non-random component from the time tagged signal received from the single-photon counting module, and subsequently filtering non-obvious patterns from the signal to extract true quantum random numbers.

A method of quantum random number generator is described. The method of quantum random number generator comprising the steps of (a) generating and detecting a photon, by a photon generator and detection module; (b) time tagging the photon, by a single-photon counting module; and (c) eliminating non-random component from the signal received from the single-photon counting module, and subsequently filtering non-obvious pattern, through a processing module, to extract true quantum random numbers.
In another embodiment, a method of quantum random number generator is described, the method comprising the steps of (a) eliminating non-random component, by a processing module, from a signal; and (b) filtering non-obvious pattern from the signal, by the processing module, to extract true quantum random numbers.

BRIEF DESCRIPTION OF DRAWINGS
Reference will be made to embodiments of the invention, example of which may be illustrated in the accompanying figure(s). These figure(s) are intended to be illustrative, not limiting. Although the invention is generally described in the context of these embodiments, it should be understood that it is not intended to limit the scope of the invention to these particular embodiments.
Figure 1 shows a system of quantum random number generator according to an embodiment of present invention;
Figure 2 shows a processing of a single-photon counting module and a processing module according to an embodiment of present invention; and
Figures 3a-3e shows a simulated signal and processing thereon by a processing module according to an embodiment of present invention:
a) Figure 3a shows a graph of a simulated original signal and a simulated noise signal according to an embodiment of the present invention;
b) Figure 3b shows a graph of trend component of the simulated noise signal of Figure 3a according to an embodiment of present invention;
c) Figure 3c shows a graph of seasonality component of the simulated noise signal of Figure 3a according to an embodiment of present invention;
d) Figure 3d shows a graph of residual component of the simulated noise signal of Figure 3a according to an embodiment of present invention; and
e) Figure 3e shows a graph of an input signal to the neural network model, and train and test data that were used to validate the neural network model reproduced post training of the neural network model according to an embodiment of present invention.
DETAILED DESCRIPTION OF THE INVENTION

A system of quantum random number generator is described. The system comprising a photon generator and detection module for generation and detection of photon, a single-photon counting module, connected to the photon generator and detection module, for time tagging the detected photon, and a processing module connected to the single-photon counting module for eliminating non-random component from the time tagged signal received from the single-photon counting module, and subsequently filtering non-obvious patterns from the signal to extract true quantum random numbers.

The photon generator and detection module comprises a laser source for generating photons, an optical attenuator to lower the intensity of the laser beam to the level of a single-photon, at least one linear polarizer for polarizing incoming photons from the attenuated laser source, at least one polarizing beam splitter for randomly transmitting or reflecting photons, and at least one detector for detection of photons.

The laser source includes a semiconductor laser diode as photon source. The wavelength of the laser is based on the spectral response of the detectors. A suitable power supply is designed to control the intensity of the laser.

The laser beam generated from the laser source is attenuated using neutral density filters of an optical attenuator to achieve the single photon intensity level (one photon per microsecond). The neutral density filters are used in series to control the degree of optical attenuation.

The linear polarizers linearly polarize incoming photons from the attenuated laser source. The orientation of the axis of the polarizer decides the output polarization state. The axis of the polarizer is aligned such that an equal superposition state is created in the polarization space.

The polarizing beam splitters randomly transmit or reflect photons conditioned on their polarization. The photon in an equal superposition of horizontal and vertical polarization states has an equal probability of getting transmitted or reflected at the polarizing beam splitter.

The detectors are low efficiency detectors, including but not limited to, LEDs (light emitting diode) operating in reverse bias. The detectors are set at an appropriate bias voltage. An incident photon, absorbed in the junction creates an electron-hole pair. The electric field in the junction will cause the pair to separate before they can recombine. The excited electron accelerates due to the electric field in the junction, creating another electron-hole pair which is subject to the same process leading to an avalanche of electrons. This avalanche process produces detectable current pulses every time a photon is incident and is called an avalanche current. The avalanche current can be measured using quenching circuits.

The single-photon counting module comprises an electronics unit which time tags the detected photons. The amplitude of the incoming current pulse from the detectors is compared with a set threshold. Pulses higher than the threshold denote detection of single photon and is recorded as 1, otherwise it is recorded as 0. Thus, small amplitude noise pulses are eliminated. The detectors are also prone to other spurious effects and noise. The processing module is configured to identify, characterize, and remove these spurious effects from the signals received from the single-photon counting module.

The processing module is connected to the single-photon counting module for eliminating non-random component from the signal received from the single-photon counting module, and thereafter filtering the non-obvious patterns from the signal to extract true quantum random number.

The processing module comprises an elimination unit configured to identify and eliminate non-random component including trend and seasonality from the signal received from the single-photon counting module; a signal processing unit configured to process the residual signal received from the elimination unit, wherein the modulus of residual signal is computed and the signal is rounded to 0 or 1; a machine learning unit configured to receive the processed signal from the signal processing unit and predict the signal having non-obvious pattern using neural network model; and a processing unit, connected to the signal processing unit and the machine learning unit, for filtering the non-obvious pattern from the signal received from the signal processing unit and extracting true quantum random number, wherein the processing unit is configured to:
a) receive the signal from the signal processing unit and compare it with the predicted signal received from the machine learning unit to filter non-obvious pattern, wherein the part of signal which overlaps with each other are discarded to obtain true single-photon signal;
b) extract the data from the filtered signal in form of 0 and 1; and
c) store the data (true quantum random numbers) in a storage unit of the processing unit.

The elimination unit is configured to identify and eliminate the non-random component from the signal (S1) received from the single-photon counting module. The signal (S1) can be decomposed into trend, seasonal, and residual components. Wherein, the trend shows the tendency of the signal to increase or decrease in magnitude over a long period of time, and the seasonal component captures the periodicity in the signal. The trend and seasonal components of the signal (S1) are not random and therefore, are being removed.

In an embodiment, the following steps describes identification and removal (elimination) of trend and seasonal components from signal (S1), received from the single-photon counting module, by the elimination unit:

a) Elimination of trend component:
Detrending is accomplished by subtracting the moving average (MA) computed using a window size such that the trend curve is smooth. The moving average (MA) for a given window is calculated using the following formula:

where is the total number of data points in the given window (window size), and is the value of the time series at time , wherein 1 = = .

The sequence of MAs results in the trend curve. This trend curve is subtracted from signal (S1) and the resulting signal (S1') is further processed to remove seasonality component.

The detrending potentially eliminates trend due to known classical sources of noise, such as bias in the beam-splitter, polarization drift, detector dark counts, etc.

b) Elimination of seasonal component:
The seasonal component is periodic, the periodicity in the discrete signal can be identified using the Discrete Fourier Transform (DFT). The seasonality curve is the summation of sines of all the prominent frequencies multiplied with their amplitudes as shown below:

where is the frequency and is the time variable. The seasonality curve is subtracted from signal (S1') to obtain signal (S2), the residual component.

In another embodiment, the decomposition of signal (S1) into seasonal, trend, and residual components can be performed, including but not limited to, using Seasonal and Trend decomposition using locally estimated scatterplot smoothing algorithm.

The Residual component (S2) of signal (S1) received from the elimination unit is fed into the signal processing unit. The signal processing unit computes the modulus of the signal and rounds it to 0 or 1, the signal processing unit outputs the processed signal (S3). Thereafter, the processed signal (S3) is sent to the machine learning unit of the processing module. The machine learning unit comprises a neural network model trained to the predict the signal having non-obvious pattern in the input signal. The predicted signal (S4) of the neural network module captures non-obvious pattern. The predicted signal (S4) of the neural network module is subtracted from the processed signal (S3) to obtain true single-photon signal (S) by a processing unit of the machine learning unit.

The machine learning unit comprises the neural network model (NN) for identifying the non-obvious patterns in the input signal. The method of preparing and training the NN to predict non-obvious pattern in the input signal comprises:

a) Creating Train and Test data
The input signal is split into train and test data sets in a specified ratio, for example 70:30. The train data is used to train the model whereas the test data is used to validate the model.

b) Choosing a certain time window (w)
The time window w has to be treated as a parameter. A value of w for which the NN model produces the best result is chosen. The best result can be decided based on a metric such as mean squared error calculated between the actual data and the data predicted by the NN.

c) Preparing the data
The train data is stacked into one dimensional arrays of size w so that it can be fed into the NN model.

d) Defining the NN model
The NN model consists of an input layer, hidden layers and an output layer. The input layer receives the data, the hidden layers transform the input data (often over multiple layers) to produce the output which is obtained in the output layer. Training the NN involves mainly two steps. The first being forward feed wherein the data traverses the input layer to the output layer. The second being the back propagation, wherein known data is used to minimize the difference between the actual and the predicted data. These steps are repeated a certain number of times and each run is called an epoch.

The present NN is defined as follows:
i. The input layer has w neurons;
ii. The number of hidden layers is a hyper-parameter and depends on the degree of non-linearity of the signal; and
iii. the output layer consists of a single neuron, for example, the (w + 1)th data point is predicted using w data points. However, the output layer can have more than one neuron.

e) Model Finalisation
The NN model is finalised by having the model predict the train and test data. Models prone to large errors and conditions such as overfitting (where the model predicts only the training data accurately) are discarded. Parameters like w, number of hidden layers, number of epochs etc. are tuned to obtain the best results.

The NN is retrained periodically to account for changes in the process due to variations in the optical setup caused by fluctuations in the environment.

An example of generating a six digit one time password (OTP) by a processing unit of a processing module is described. The string S (true quantum random number) stored in binary digits in a storage unit of the processing unit is divided into sub-strings of 8 bits. For instance, the first string obtained is 10011100, it is converted to corresponding decimal number i.e. 156. Similarly, for instance, the second string is 11100110, it is converted to the corresponding decimal value of 230. Thereafter, the obtained decimal values are concatenated to form a six digit quantum random number “156230” which can be used as an OTP.

In another example, the true quantum random numbers stored in the storage unit in binary digits can be directly used as random numbers.

A method of quantum random number generator is described. The method comprising the steps of:
a) generating and detecting a photon, by a photon generator and detection module;
b) time tagging the photon, by a single-photon counting module; and
c) eliminating non-random component from the signal received from the single-photon counting module, and subsequently filtering non-obvious pattern, through a processing module, to extract true quantum random numbers.

The step (a) of generating and detecting the photon includes the step of:
a) generating a photon by a laser source;
b) lowering the intensity of laser beam to the single-photon level, by an optical attenuator;
c) polarizing incoming photons, by a linear polarizer, from the attenuated laser source;
d) randomly transmitting or reflecting photons, by a polarizing beam splitter; and
e) detecting a single-photon, by a detector.

The step (c) of eliminating the non-random component from the signal received from the single-photon counting module, and subsequently filtering the non-obvious pattern, through a processing module, to extract true quantum random numbers, includes the steps of:
a) identifying and eliminating, by an elimination unit, non-random component, including trend and seasonality, from the signal received from the single-photon counting module;
b) processing residual signal received from the elimination unit, by a signal processing unit, wherein the modulus of residual signal is computed and the signal is rounded to 0 or 1;
c) predicting the signal having non-obvious pattern, by a machine learning unit, using neural network model; and
d) processing, by a processing unit, to filter non-obvious pattern from the signal received from the signal processing unit and extract true quantum random number, wherein the processing by the processing unit comprises the steps of:
i. receiving the signal from the signal processing unit and comparing it with the predicted signal received from the neural network model of the machine learning unit to filter non-obvious pattern, wherein the part of signal which overlaps with each other are discarded to obtain true single-photon signal;
ii. extracting the data from the filtered signal in form of 0 and 1; and
iii. storing the data (true quantum random numbers) in a storage unit of the processing unit.

The machine learning unit comprises neural network model for predicting the signal to filter non-obvious pattern, the neural network model is developed using the steps:
a) preparing train and test data, wherein the signal received to the machine learning unit is split into train and test data sets in a specified ratio;
b) choosing a certain time window (w), wherein the time window (w) is treated as a parameter, the value of w for which the neural network model produces the best result is chosen;
c) preparing the data, wherein the train data is stacked into one dimensional arrays of size w so that it can be fed into the neural network model;
d) defining the neural network model, wherein neural network model consists of an input layer, a hidden layer, and an output layer; and
e) finalizing the neural network model, wherein the neural network model which has minimum error is finalized, an appropriate error metric is chosen to calculate the error between the original signal and the predicted signal by the NN post training.

The step of filtering non-obvious pattern from the signal includes, but is not limited to, thermal noise, dark counts (false positives) in the detector(s), and jitters in the optical circuit due to spurious mechanical vibrations.

In another embodiment, a method of quantum random number generator is described, the method comprising the steps of (a) eliminating non-random component, by a processing module, from a signal; and (b) filtering non-obvious pattern from the signal, by the processing module, to extract true quantum random numbers.

The subject matter is now described with reference to the drawings, wherein like reference numerals are used to refer to like elements throughout. In the following description, for purpose of explanation, numerous specific details are set forth in order to provide a thorough understanding of the claimed subject matter. It may be evident however, that such matter can be practiced with these specific details. In other instances, well-known structures as shown in diagram form in order to facilitate describing the invention.

Referring Figure 1, a system (100) of quantum random number generator is shown. As shown, the system (100) comprises a photon generator and detection module (110) for generation and detection of photon, a single-photon counting module (120), connected to the photon generator and detection module (110), configured to time tag the detected photon, and a processing module (130), connected to the single-photon counting module (120), for eliminating non-random component from the time tagged signal received from the single-photon counting module (120), and subsequently filtering non-obvious patterns from the signal to extract true quantum random numbers.

As shown in Figure 1, the photon generator and detection module (110) comprises a laser source (111) for generating photon, an optical attenuator (112) for lowering the intensity of laser beam to the single-photon level, a polarizer (1131, 1132, 1133) for polarizing incoming photons from the attenuated laser source, a polarizing beam splitter (1141, 1142, 1143) for randomly transmitting or reflecting photons, and a detector (1151, 1152, 1153, 1154) for detection of single-photon.

As shown in Figure 1, the detectors (1151, 1152, 1153, 1154) are low efficiency single-photon detector which have low cost. The detector (1151, 1152, 1153, 1154) drastically reduce the cost of the system, thereby resulting in affordability of the system to be utilized in quantum random number generator.

As shown in Figure 1, attenuated laser source is obtained by passing the laser beam from the laser source (111) through the optical attenuator (112). The photons from the attenuated laser source, which are, for instance, initially vertically polarized, are passed through a polarizer (1131) oriented at 45 degrees creating a superposition state. These photons are then incident on the polarizing beam splitter (1141) and have 50% probability of getting transmitted or reflected. The photons that are transmitted are again put into a superposition state by polarizer (1132). As described earlier, the photons get either transmitted or reflected at the polarizing beam splitter (1142), which are accordingly detected at the detector (1151) or detector (1152). Similarly, the reflected photons are detected at the detector (1153) or the detector (1154).

As shown in Figure 1, the signals measured at the detectors (1151, 1152, 1153, 1154) may comprise of false detection events arising due to dark counts, multiple photon detections and classical randomness due to noise in the detectors such as thermal noise. In order to extract pure quantum randomness, the above mentioned unwanted components are eliminated from the recorded signal. Dark counts are accounted for by carefully characterising the detectors. Further, due to use of low intensity coherent source, it is possible for two photons to simultaneously and independently traverse the two arms of the setup, causing detection at the detector (1151) or detector (1152) and detector (1153) or detector (1154). The multi-photon detection events cause both the detectors to click simultaneously (within a specified time window). These events, characterized by coincident detections at the detector (1151) and the detector (1152) or the detector (1153) and the detector (1154) are discarded. After eliminating coincidence counts, a signal is obtained which comprises of true single-photon detection events, classical randomness (arising from dark counts) and other systematic noise.

Referring Figure 2, a processing (200) of a single-photon counting module (210) and a processing module (220) is shown. As shown, the signal (data) received at the single-photon counting module (210) through detectors (not shown) is time tagged. The time tagged signal (data) from the single-photon counting module is sent to the processing module (220) for eliminating non-random components from the time tagged signal (data) received from the single-photon counting module (210), and subsequently filtering non-obvious pattern from the signal to extract true quantum random numbers.

As shown in Figure 2, the processing module (220) comprises an elimination unit (221), a signal processing unit (222), a machine learning unit (223), and a processing unit (224). The processing module (220) on receipt of the signal (data) from the single-photon counting module (210) first eliminates the non-random components, including trend and seasonality, from the signal (data) through the elimination unit (221), subsequently, the residual component of the signal (data) is sent to the signal processing unit (222). The signal processing unit (222) on receipt of the signal calculates the modulus of the signal and the signal is rounded to 0 or 1. Explicitly, values below 0.5 are rounded to 0, and 0.5 and upwards are be rounded to 1.

As shown in Figure 2, the processed signal is sent to the machine learning unit (223) which comprises a neural network model (2231) to predict the signal having non-obvious pattern. The predicted signal from the machine learning unit (222) is sent to the processing unit (224). The processing unit (224) on receipt of predicted signal from the machine learning unit (223) compares the predicted signal (data) with the signal (data) received from the signal processing unit (222), and the part (instances) of signal which overlaps with each other are discarded to obtain true single-photon signal.

As shown in Figure 2, the processing unit (224) after filtering the noise from the signal, is configured to extract the data from the filtered signal in form of 0 and 1, and store the data (true quantum random numbers) in set of string of 1 and 0 in a storage unit (not shown). The true quantum random number stored in the storage unit can be directly used as random numbers in binary form or the processing unit is configured to fetch the binary digits from the storage unit based on number of digits of random number, and thereafter, converting the binary value to any number number system, including, decimal number system to generate the random number.

As shown in Figure 2, the neural network model (2231) of the machine learning unit (223) comprises an input layer, a hidden layer, and an output layer.

Figures 3a-3e shows a simulated signal and processing thereon by a processing module according to an embodiment of present invention.

Figure 3a shows a graph (300) of a simulated original signal (310) and a simulated noise signal (320) according to an embodiment of the present invention.

The signals were simulated using the following steps:

a. A sequence of random bits were generated. Bit '0' was mapped to "no photon detected" and bit '1' was mapped to "photon(s) were detected" events respectively;

b. The string was iterated over to assign a pulse height (chosen as 0.25 corresponding to zero and 0.75 corresponding to one);

c. The noise component was added to the sequence of pulse heights; and

d. The modulus of this sequence was rounded to the nearest integer (zero or one in the present case 0.5 as the threshold).

The noise model comprises of:
a. A periodic noise with frequency 75Hz; and

b. A Poisson noise with rate parameter set to 8.

As shown in Figure 3a, the simulated noise signal (320) comprises of trend, seasonality, and residual components. The X-axis (340) is discrete relative time and the Y-axis (330) is the amplitude.

The simulated noise signal (320) of Figure 3a is decomposed, for example, using the Seasonal and Trend decomposition using locally estimated scatterplot smoothing algorithm, by an elimination unit (not shown) of the processing module (not shown). The trend component of the signal (320) is shown in Figure 3b, the seasonality component of the signal (320) is shown in Figure 3c, and the residual component of the signal (320) is shown in Figure 3d.

Figure 3b shows a graph (400) of trend component (430) of the simulated noise signal (320) of Figure 3a according to an embodiment of present invention. The X-axis (420) is discrete relative time and the Y-axis (410) is the amplitude.

Figure 3c, shows a graph (500) of seasonality component (530) of the simulated noise signal (320) of Figure 3a according to an embodiment of present invention. The X-axis (520) is discrete relative time and the Y-axis (510) is the amplitude.

Figure 3d shows a graph (600) of residual component (630) of the simulated noise signal (320) of Figure 3a according to an embodiment of present invention. The X-axis (620) is discrete relative time and the Y-axis (610) is the amplitude.

The residual component (630) of the signal (320) obtained after decomposition from the elimination unit (not shown) is sent to a signal processing unit (not shown). The signal processing unit (not shown) calculates the modulus of the signal and rounds the signal to 0 or 1. The signal after being processed in the signal processing unit (not shown) is sent to a machine learning unit (not shown) of the processing module (not shown), wherein the machine learning unit (not shown) comprises a neural network model (not shown) configured to output the non-obvious patterns. The neural network model (not shown) with three neurons in the input layer, four middle layers and an output layer with a single neuron is considered. The signal i.e. residual component received from the signal processing unit (not shown) is divided in train and test data in the ratio 67:33. The model is trained on the train data with 150 epochs. The model was tuned for optimal performance.

The metric chosen to evaluate the model is the Root Mean Square Error (RMSE). The train and test scores were 0.35 and 0.37 respectively. The results of processing of signal is shown in Figure 3e.

Figure 3e, shows a graph (700) of the input signal (720) to the neural network model (not shown), and train and test data that were used to validate the neural network model (not shown) reproduced post training of the neural network model according to an embodiment of present invention. The solid line (720) corresponds to input signal i.e. residual component (630) after being processed from the signal processing unit (not shown). The dashed line (710) and the dotted line (730) correspond to the predicted train and test data respectively. The X-axis (750) is discrete relative time and the Y-axis (740) is the amplitude.

INDUSTRIAL APPLICABILITY
The system and method of quantum random number generator described herein utilizes low efficiency detectors, which are highly cost effective, for detection of photons thereby drastically reducing the cost and making it affordable for use in low budget applications.

In addition, the system and method utilizes processing module to remove non-random components, and filter non-obvious pattern, if any, which enhances the capability and efficiency of the system and method to produce true quantum random numbers.

The foregoing description of the invention has been set merely to illustrate the invention and is not intended to be limiting. Since modifications of the disclosed embodiments incorporating the substance of the invention may occur to person skilled in the art, the invention should be construed to include everything within the scope of the disclosure.