{
  "id": 251742,
  "title": "Questions about SNR",
  "url": "/competitions/g2net-gravitational-wave-detection/discussion/251742",
  "author_name": "",
  "post_date": "2021-07-08T15:40:57.946183800Z",
  "votes": 14,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I was just reading the section about SNR on data page</p>\n<blockquote>\n  <p>The integrated signal-to noise ratio (SNR) is classically the most informative measure of how detectable a signal is and a typical level of detectability is when this integrated SNR exceeds ~8. This shouldn't confused with the instantaneous SNR - the factor by which the signal rises above the noise - and in nearly all cases the (unlike the first gravitational wave detection GW150914) these signals are not visible by eye in the time series. </p>\n</blockquote>\n<p>This may sound a bit ignorant, but so far i just followed the normal workflow (wave -&gt; image -&gt; cnn) like most kernel did without even thought about SNR. So Im just trying to understand how would this be relevant/helpful to our prediction task. </p>\n<p>From my understanding about SNR, it's simply the power ratio between signal and noise, and the higher the relatively stronger signal present, and should also be relatively easier to detect. I don't fully understand instantaneous SNR vs integrated SNR, integrated SNR sounds like the SNR is computed over a region (i guess hence the integration) and instantaneous SNR is just pointwise calculation somehow?</p>\n<blockquote>\n  <p>these signals are not visible by eye in the time series. </p>\n</blockquote>\n<p>if the wave has integrated SNR &gt; 8 but not visible by eyes, does that mean the signals are very scattered across different time point? otherwise i dont understand how could there is integrrated signals but not visible by eyes. </p>\n<p>Are our data generated all have SNR=8 or it can be anything reasonable at &gt; 8?</p>\n<p>Where does the noise come from?</p>\n<ol>\n<li>i read that the measurement is super sensitive, so a truck going nearby might be registered?</li>\n<li>the inteferometer relies on laser, and laser itself does not have all photons focused to exactly one place, so noise could also come from the natural distribution from the photons from laser?</li>\n</ol>",
  "messages": [
    {
      "id": "1381071",
      "postDate": "07/08/2021 15:40:57",
      "content": "<p>I was just reading the section about SNR on data page</p>\n<blockquote>\n  <p>The integrated signal-to noise ratio (SNR) is classically the most informative measure of how detectable a signal is and a typical level of detectability is when this integrated SNR exceeds ~8. This shouldn't confused with the instantaneous SNR - the factor by which the signal rises above the noise - and in nearly all cases the (unlike the first gravitational wave detection GW150914) these signals are not visible by eye in the time series. </p>\n</blockquote>\n<p>This may sound a bit ignorant, but so far i just followed the normal workflow (wave -&gt; image -&gt; cnn) like most kernel did without even thought about SNR. So Im just trying to understand how would this be relevant/helpful to our prediction task. </p>\n<p>From my understanding about SNR, it's simply the power ratio between signal and noise, and the higher the relatively stronger signal present, and should also be relatively easier to detect. I don't fully understand instantaneous SNR vs integrated SNR, integrated SNR sounds like the SNR is computed over a region (i guess hence the integration) and instantaneous SNR is just pointwise calculation somehow?</p>\n<blockquote>\n  <p>these signals are not visible by eye in the time series. </p>\n</blockquote>\n<p>if the wave has integrated SNR &gt; 8 but not visible by eyes, does that mean the signals are very scattered across different time point? otherwise i dont understand how could there is integrrated signals but not visible by eyes. </p>\n<p>Are our data generated all have SNR=8 or it can be anything reasonable at &gt; 8?</p>\n<p>Where does the noise come from?</p>\n<ol>\n<li>i read that the measurement is super sensitive, so a truck going nearby might be registered?</li>\n<li>the inteferometer relies on laser, and laser itself does not have all photons focused to exactly one place, so noise could also come from the natural distribution from the photons from laser?</li>\n</ol>",
      "rawMarkdown": "I was just reading the section about SNR on data page\n> The integrated signal-to noise ratio (SNR) is classically the most informative measure of how detectable a signal is and a typical level of detectability is when this integrated SNR exceeds ~8. This shouldn't confused with the instantaneous SNR - the factor by which the signal rises above the noise - and in nearly all cases the (unlike the first gravitational wave detection GW150914) these signals are not visible by eye in the time series. \n\nThis may sound a bit ignorant, but so far i just followed the normal workflow (wave -> image -> cnn) like most kernel did without even thought about SNR. So Im just trying to understand how would this be relevant/helpful to our prediction task. \n\nFrom my understanding about SNR, it's simply the power ratio between signal and noise, and the higher the relatively stronger signal present, and should also be relatively easier to detect. I don't fully understand instantaneous SNR vs integrated SNR, integrated SNR sounds like the SNR is computed over a region (i guess hence the integration) and instantaneous SNR is just pointwise calculation somehow?\n\n> these signals are not visible by eye in the time series. \n\nif the wave has integrated SNR > 8 but not visible by eyes, does that mean the signals are very scattered across different time point? otherwise i dont understand how could there is integrrated signals but not visible by eyes. \n\nAre our data generated all have SNR=8 or it can be anything reasonable at > 8?\n\nWhere does the noise come from?\n1. i read that the measurement is super sensitive, so a truck going nearby might be registered?\n2. the inteferometer relies on laser, and laser itself does not have all photons focused to exactly one place, so noise could also come from the natural distribution from the photons from laser?",
      "votes": null
    },
    {
      "id": "1382624",
      "postDate": "07/10/2021 04:37:53",
      "content": "<p>Same question for me, upvoted and will follow</p>",
      "rawMarkdown": "Same question for me, upvoted and will follow",
      "votes": null
    },
    {
      "id": "1383066",
      "postDate": "07/10/2021 13:27:39",
      "content": "<p>Hi,<br>\nHere are my two cents;<br>\nI also had a same feeling at first, and now I think that the statements below says mainly two things;</p>\n<p>===begin excerpt in data section===<br>\nThe integrated signal-to noise ratio (SNR) is classically the most informative measure of how detectable a signal is and a typical level of detectability is when this integrated SNR exceeds ~8. This shouldn't confused with the instantaneous SNR - the factor by which the signal rises above the noise - and in nearly all cases the (unlike the first gravitational wave detection GW150914) these signals are not visible by eye in the time series.<br>\n===end of excerpt===</p>\n<ol>\n<li>integrated SNR and instantaneous SNR are different, thus shouldn't be confused<br>\n(SNR value of 8 should not be an essential message, I suppose).<br>\nAND</li>\n<li>signals are not visible for almost all the cases in the provided dataset</li>\n</ol>\n<p>Below are details;</p>\n<ol>\n<li>integrated SNR and instantaneous SNR are different, thus shouldn't be confused.</li>\n</ol>\n<p>For integrated SNR, I agree with you, this should be the SNR over a certain period of time (or band width).<br>\nFor instantaneous SNR, I suppose this SNR is defined for each time sample (analogous to \"instantaneous frequency\" in signal processing-wise).</p>\n<p>I think that the author implicitly says that this instantaneous SNR is more suitable than the integrated SNR for this data. This is probably because the signal level of GW signal is time-varying and duration is not constant, hence the integrated SNR is difficult to apply (the integrated SNR value changes according to the time width).</p>\n<ol>\n<li>signals are not visible for almost all the cases in the provided dataset.<br>\nWhatever the definition of SNR is applied, the signals are not very visible in raw data.<br>\nAnd this is the very message of the author for this dataset - we cannot assume that the GW data is visible at least in raw data -&gt; challenging point of this competition :-)</li>\n</ol>\n<p>Comments to improve this my understanding are very much appreciated !<br>\nBest regards,<br>\n    Shinji</p>",
      "rawMarkdown": "Hi,\nHere are my two cents;\nI also had a same feeling at first, and now I think that the statements below says mainly two things;\n\n===begin excerpt in data section===\nThe integrated signal-to noise ratio (SNR) is classically the most informative measure of how detectable a signal is and a typical level of detectability is when this integrated SNR exceeds ~8. This shouldn't confused with the instantaneous SNR - the factor by which the signal rises above the noise - and in nearly all cases the (unlike the first gravitational wave detection GW150914) these signals are not visible by eye in the time series.\n===end of excerpt===\n\n1. integrated SNR and instantaneous SNR are different, thus shouldn't be confused\n(SNR value of 8 should not be an essential message, I suppose).\nAND\n2. signals are not visible for almost all the cases in the provided dataset\n\nBelow are details;\n1. integrated SNR and instantaneous SNR are different, thus shouldn't be confused.\n\nFor integrated SNR, I agree with you, this should be the SNR over a certain period of time (or band width).\nFor instantaneous SNR, I suppose this SNR is defined for each time sample (analogous to \"instantaneous frequency\" in signal processing-wise).\n\nI think that the author implicitly says that this instantaneous SNR is more suitable than the integrated SNR for this data. This is probably because the signal level of GW signal is time-varying and duration is not constant, hence the integrated SNR is difficult to apply (the integrated SNR value changes according to the time width).\n\n2. signals are not visible for almost all the cases in the provided dataset.\nWhatever the definition of SNR is applied, the signals are not very visible in raw data.\nAnd this is the very message of the author for this dataset - we cannot assume that the GW data is visible at least in raw data -> challenging point of this competition :-)\n\n\nComments to improve this my understanding are very much appreciated !\nBest regards,\n    Shinji",
      "votes": null
    },
    {
      "id": "1386185",
      "postDate": "07/13/2021 10:11:25",
      "content": "<p>Hi Sam?,</p>\n<p>These are all great questions. Let me see if I can help clarify things. Firstly, I appreciate that you don't need to know the SNR to classify the data. Our intent when including the text on SNR was to give a hint that you shouldn't expect to be able to see the vast majority of the signals by-eye - although some may be strong enough to see.</p>\n<p>Your understanding about the general meaning of integrated SNR and instantaneous SNR is correct. The integrated SNR is essentially accumulated SNR over the duration of the signal, and as you can tell from the published detections, these signals last for a finite amount of time (many time samples).</p>\n<p>I can't really tell you too much about the distribution of integrated SNRs present in the data other than to say that there is a broad range.</p>\n<p>Where the noise comes from is a very good question. Have a look at this paper <a href=\"https://link.springer.com/article/10.12942/lrr-2011-5\" target=\"_blank\">living review</a> for a nice summary of the noise sources in a real gravitational wave detector. The data that you have in the challenge is simulated to model signals in gravitational wave noise that represent these nearly-complete set of expected noise sources.</p>\n<p>I hope this helps.</p>\n<p>Chris </p>",
      "rawMarkdown": "Hi Sam?,\n\nThese are all great questions. Let me see if I can help clarify things. Firstly, I appreciate that you don't need to know the SNR to classify the data. Our intent when including the text on SNR was to give a hint that you shouldn't expect to be able to see the vast majority of the signals by-eye - although some may be strong enough to see.\n\nYour understanding about the general meaning of integrated SNR and instantaneous SNR is correct. The integrated SNR is essentially accumulated SNR over the duration of the signal, and as you can tell from the published detections, these signals last for a finite amount of time (many time samples).\n\nI can't really tell you too much about the distribution of integrated SNRs present in the data other than to say that there is a broad range.\n\nWhere the noise comes from is a very good question. Have a look at this paper [living review](https://link.springer.com/article/10.12942/lrr-2011-5) for a nice summary of the noise sources in a real gravitational wave detector. The data that you have in the challenge is simulated to model signals in gravitational wave noise that represent these nearly-complete set of expected noise sources.\n\nI hope this helps.\n\nChris",
      "votes": null
    },
    {
      "id": "1563274",
      "postDate": "10/28/2021 06:53:02",
      "content": "<p>Hey,</p>\n<p>Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey <a href=\"https://forms.gle/QP9L16niPexozyhu5\" target=\"_blank\">https://forms.gle/QP9L16niPexozyhu5</a>.</p>\n<p>Thank you all,</p>\n<p>Regards,<br>\nChris</p>",
      "rawMarkdown": "Hey,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1382624,
      "author_name": "superchenhao",
      "author_url": "",
      "post_date": "07/10/2021 04:37:53",
      "content": "<p>Same question for me, upvoted and will follow</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1383066,
      "author_name": "shinjiyoneshima",
      "author_url": "",
      "post_date": "07/10/2021 13:27:39",
      "content": "<p>Hi,<br>\nHere are my two cents;<br>\nI also had a same feeling at first, and now I think that the statements below says mainly two things;</p>\n<p>===begin excerpt in data section===<br>\nThe integrated signal-to noise ratio (SNR) is classically the most informative measure of how detectable a signal is and a typical level of detectability is when this integrated SNR exceeds ~8. This shouldn't confused with the instantaneous SNR - the factor by which the signal rises above the noise - and in nearly all cases the (unlike the first gravitational wave detection GW150914) these signals are not visible by eye in the time series.<br>\n===end of excerpt===</p>\n<ol>\n<li>integrated SNR and instantaneous SNR are different, thus shouldn't be confused<br>\n(SNR value of 8 should not be an essential message, I suppose).<br>\nAND</li>\n<li>signals are not visible for almost all the cases in the provided dataset</li>\n</ol>\n<p>Below are details;</p>\n<ol>\n<li>integrated SNR and instantaneous SNR are different, thus shouldn't be confused.</li>\n</ol>\n<p>For integrated SNR, I agree with you, this should be the SNR over a certain period of time (or band width).<br>\nFor instantaneous SNR, I suppose this SNR is defined for each time sample (analogous to \"instantaneous frequency\" in signal processing-wise).</p>\n<p>I think that the author implicitly says that this instantaneous SNR is more suitable than the integrated SNR for this data. This is probably because the signal level of GW signal is time-varying and duration is not constant, hence the integrated SNR is difficult to apply (the integrated SNR value changes according to the time width).</p>\n<ol>\n<li>signals are not visible for almost all the cases in the provided dataset.<br>\nWhatever the definition of SNR is applied, the signals are not very visible in raw data.<br>\nAnd this is the very message of the author for this dataset - we cannot assume that the GW data is visible at least in raw data -&gt; challenging point of this competition :-)</li>\n</ol>\n<p>Comments to improve this my understanding are very much appreciated !<br>\nBest regards,<br>\n    Shinji</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1386185,
      "author_name": "bayeswolf",
      "author_url": "",
      "post_date": "07/13/2021 10:11:25",
      "content": "<p>Hi Sam?,</p>\n<p>These are all great questions. Let me see if I can help clarify things. Firstly, I appreciate that you don't need to know the SNR to classify the data. Our intent when including the text on SNR was to give a hint that you shouldn't expect to be able to see the vast majority of the signals by-eye - although some may be strong enough to see.</p>\n<p>Your understanding about the general meaning of integrated SNR and instantaneous SNR is correct. The integrated SNR is essentially accumulated SNR over the duration of the signal, and as you can tell from the published detections, these signals last for a finite amount of time (many time samples).</p>\n<p>I can't really tell you too much about the distribution of integrated SNRs present in the data other than to say that there is a broad range.</p>\n<p>Where the noise comes from is a very good question. Have a look at this paper <a href=\"https://link.springer.com/article/10.12942/lrr-2011-5\" target=\"_blank\">living review</a> for a nice summary of the noise sources in a real gravitational wave detector. The data that you have in the challenge is simulated to model signals in gravitational wave noise that represent these nearly-complete set of expected noise sources.</p>\n<p>I hope this helps.</p>\n<p>Chris </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1563274,
      "author_name": "zerafachris",
      "author_url": "",
      "post_date": "10/28/2021 06:53:02",
      "content": "<p>Hey,</p>\n<p>Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey <a href=\"https://forms.gle/QP9L16niPexozyhu5\" target=\"_blank\">https://forms.gle/QP9L16niPexozyhu5</a>.</p>\n<p>Thank you all,</p>\n<p>Regards,<br>\nChris</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1381071": "I was just reading the section about SNR on data page\n> The integrated signal-to noise ratio (SNR) is classically the most informative measure of how detectable a signal is and a typical level of detectability is when this integrated SNR exceeds ~8. This shouldn't confused with the instantaneous SNR - the factor by which the signal rises above the noise - and in nearly all cases the (unlike the first gravitational wave detection GW150914) these signals are not visible by eye in the time series. \n\nThis may sound a bit ignorant, but so far i just followed the normal workflow (wave -> image -> cnn) like most kernel did without even thought about SNR. So Im just trying to understand how would this be relevant/helpful to our prediction task. \n\nFrom my understanding about SNR, it's simply the power ratio between signal and noise, and the higher the relatively stronger signal present, and should also be relatively easier to detect. I don't fully understand instantaneous SNR vs integrated SNR, integrated SNR sounds like the SNR is computed over a region (i guess hence the integration) and instantaneous SNR is just pointwise calculation somehow?\n\n> these signals are not visible by eye in the time series. \n\nif the wave has integrated SNR > 8 but not visible by eyes, does that mean the signals are very scattered across different time point? otherwise i dont understand how could there is integrrated signals but not visible by eyes. \n\nAre our data generated all have SNR=8 or it can be anything reasonable at > 8?\n\nWhere does the noise come from?\n1. i read that the measurement is super sensitive, so a truck going nearby might be registered?\n2. the inteferometer relies on laser, and laser itself does not have all photons focused to exactly one place, so noise could also come from the natural distribution from the photons from laser?",
    "1382624": "Same question for me, upvoted and will follow",
    "1383066": "Hi,\nHere are my two cents;\nI also had a same feeling at first, and now I think that the statements below says mainly two things;\n\n===begin excerpt in data section===\nThe integrated signal-to noise ratio (SNR) is classically the most informative measure of how detectable a signal is and a typical level of detectability is when this integrated SNR exceeds ~8. This shouldn't confused with the instantaneous SNR - the factor by which the signal rises above the noise - and in nearly all cases the (unlike the first gravitational wave detection GW150914) these signals are not visible by eye in the time series.\n===end of excerpt===\n\n1. integrated SNR and instantaneous SNR are different, thus shouldn't be confused\n(SNR value of 8 should not be an essential message, I suppose).\nAND\n2. signals are not visible for almost all the cases in the provided dataset\n\nBelow are details;\n1. integrated SNR and instantaneous SNR are different, thus shouldn't be confused.\n\nFor integrated SNR, I agree with you, this should be the SNR over a certain period of time (or band width).\nFor instantaneous SNR, I suppose this SNR is defined for each time sample (analogous to \"instantaneous frequency\" in signal processing-wise).\n\nI think that the author implicitly says that this instantaneous SNR is more suitable than the integrated SNR for this data. This is probably because the signal level of GW signal is time-varying and duration is not constant, hence the integrated SNR is difficult to apply (the integrated SNR value changes according to the time width).\n\n2. signals are not visible for almost all the cases in the provided dataset.\nWhatever the definition of SNR is applied, the signals are not very visible in raw data.\nAnd this is the very message of the author for this dataset - we cannot assume that the GW data is visible at least in raw data -> challenging point of this competition :-)\n\n\nComments to improve this my understanding are very much appreciated !\nBest regards,\n    Shinji",
    "1386185": "Hi Sam?,\n\nThese are all great questions. Let me see if I can help clarify things. Firstly, I appreciate that you don't need to know the SNR to classify the data. Our intent when including the text on SNR was to give a hint that you shouldn't expect to be able to see the vast majority of the signals by-eye - although some may be strong enough to see.\n\nYour understanding about the general meaning of integrated SNR and instantaneous SNR is correct. The integrated SNR is essentially accumulated SNR over the duration of the signal, and as you can tell from the published detections, these signals last for a finite amount of time (many time samples).\n\nI can't really tell you too much about the distribution of integrated SNRs present in the data other than to say that there is a broad range.\n\nWhere the noise comes from is a very good question. Have a look at this paper [living review](https://link.springer.com/article/10.12942/lrr-2011-5) for a nice summary of the noise sources in a real gravitational wave detector. The data that you have in the challenge is simulated to model signals in gravitational wave noise that represent these nearly-complete set of expected noise sources.\n\nI hope this helps.\n\nChris",
    "1563274": "Hey,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris"
  },
  "source": "meta"
}